RYDE-WORK/lnp_ml
1
0
Fork 0
mirror of https://github.com/RYDE-WORK/lnp_ml.git synced 2026-07-21 21:22:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: Qwen SoftRAG（RAG prompt/soft token/QLoRA）+ Morgan 检索旁路与实验结果

Browse Source
This commit is contained in:
Michelle0574 2026-07-07 17:47:34 +00:00
parent 0fc72a9474
commit 22e83122e7
216 changed files with 2503 additions and 7826 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
.gitignore vendored
View File

@ -195,4 +195,7 @@ logs/
models/**/*.pt models/**/*.pt
models/**/*.png models/**/*.png
models/molt5-base/ models/molt5-base/
models/abl/ models/abl/
models/qwen2.5-7b-instruct/
models/**/*.sqlite3tests/
models/**/*.sqlite3

7
lnp_ml/modeling/layers/fusion.py
View File

@ -128,6 +128,7 @@ class ResidualConcatFusion(nn.Module):
# 零初始化门控（可学习标量），旁路初始不参与 # 零初始化门控（可学习标量），旁路初始不参与
self.g_moe = nn.Parameter(torch.zeros(())) self.g_moe = nn.Parameter(torch.zeros(()))
self.g_llm = nn.Parameter(torch.zeros(())) self.g_llm = nn.Parameter(torch.zeros(()))
self.g_retr = nn.Parameter(torch.zeros(()))
def forward( def forward(
self, self,
@ -135,9 +136,9 @@ class ResidualConcatFusion(nn.Module):
tab: torch.Tensor, tab: torch.Tensor,
f_moe: Optional[torch.Tensor] = None, f_moe: Optional[torch.Tensor] = None,
f_llm: Optional[torch.Tensor] = None, f_llm: Optional[torch.Tensor] = None,
f_retr: Optional[torch.Tensor] = None,
return_attn_weights: bool = False, return_attn_weights: bool = False,
) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]: ) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
# 只对真实 token（chem + tab）做注意力池化，旁路不参与 softmax 竞争
seq = torch.cat([chem, tab], dim=1) # [B, n_chem + n_cond, d_model] seq = torch.cat([chem, tab], dim=1) # [B, n_chem + n_cond, d_model]
pooled = self.pool(seq, return_attn_weights=return_attn_weights) pooled = self.pool(seq, return_attn_weights=return_attn_weights)
if return_attn_weights: if return_attn_weights:
@ -145,8 +146,10 @@ class ResidualConcatFusion(nn.Module):
out = pooled out = pooled
if f_moe is not None: if f_moe is not None:
out = out + self.g_moe * f_moe # 残差 + 零初始化门 out = out + self.g_moe * f_moe # 残差 + 零初始化门
if f_llm is not None: if f_llm is not None:
out = out + self.g_llm * f_llm out = out + self.g_llm * f_llm
if f_retr is not None:
out = out + self.g_retr * f_retr
return (out, attn) if return_attn_weights else out return (out, attn) if return_attn_weights else out

267
lnp_ml/modeling/layers/llm_prompt.py
View File

@ -1,4 +1,4 @@
"""LLM 分子特征分支：用 MolT5 直接编码 SMILES 文本。""" """LLM 分子特征分支：SMILES 文本 + chem'/tab soft token 的混合 prompt 编码。"""
import os import os
from typing import Dict, List, Optional from typing import Dict, List, Optional
@ -7,15 +7,20 @@ import torch
import torch.nn as nn import torch.nn as nn
# 权重默认路径，可用环境变量 MOLT5_PATH 覆盖
DEFAULT_MOLT5_PATH = os.environ.get("MOLT5_PATH", "models/molt5-base") DEFAULT_MOLT5_PATH = os.environ.get("MOLT5_PATH", "models/molt5-base")
# 检索池支持的额外多任务（除 delivery 外）
_EXTRA_TASKS = ["size", "pdi", "ee", "toxic", "biodist"]
class LLMPromptEncoder(nn.Module): class LLMPromptEncoder(nn.Module):
"""用 MolT5 编码 SMILES 文本，输出分子特征 F_llm [B, d_model]。 """用 LLM 编码分子，输出 F_llm [B, d_model]。
- 冻结 encoder 时：对每个 SMILES 缓存其句向量，避免重复前向，降方差、提速。 三种模式：
- use_lora=True 时：encoder 可训练（LoRA），不缓存。 - use_soft_prompt=True：原始 SMILES 文本(+RAG 邻居) + chem'/tab soft token，
inputs_embeds 注入，全程带梯度（配合 LoRA/QLoRA 真微调），不缓存特征。
- use_rag=True 且非 soft：旧的冻结+缓存 RAG 路径。
- 其他：冻结缓存 / 可训练 mean-pool。
""" """
def __init__( def __init__(
@ -26,33 +31,50 @@ class LLMPromptEncoder(nn.Module):
model_name_or_path: str = DEFAULT_MOLT5_PATH, model_name_or_path: str = DEFAULT_MOLT5_PATH,
freeze: bool = True, freeze: bool = True,
use_lora: bool = False, use_lora: bool = False,
use_qlora: bool = False,
lora_r: int = 8, lora_r: int = 8,
lora_alpha: int = 16, lora_alpha: int = 16,
lora_dropout: float = 0.05, lora_dropout: float = 0.05,
max_length: int = 128, max_length: int = 256,
use_rag: bool = False, use_rag: bool = False,
rag_top_k: int = 4, rag_top_k: int = 4,
use_soft_prompt: bool = False,
) -> None: ) -> None:
super().__init__() super().__init__()
from transformers import AutoTokenizer, T5EncoderModel, AutoModel from transformers import AutoTokenizer, T5EncoderModel, AutoModel
self.use_lora = use_lora self.use_lora = use_lora or use_qlora
self.use_qlora = use_qlora
self.max_length = max_length self.max_length = max_length
self.use_rag = use_rag
self.rag_top_k = rag_top_k
self.use_soft_prompt = use_soft_prompt
_name_l = model_name_or_path.lower() _name_l = model_name_or_path.lower()
_is_t5 = "t5" in _name_l _is_t5 = "t5" in _name_l
_is_qwen = "qwen" in _name_l _is_qwen = "qwen" in _name_l
self.use_rag = use_rag
self.rag_top_k = rag_top_k
self._is_qwen = _is_qwen self._is_qwen = _is_qwen
# Qwen 等大模型 tokenizer 需要 trust_remote_code
self.tokenizer = AutoTokenizer.from_pretrained( self.tokenizer = AutoTokenizer.from_pretrained(
model_name_or_path, trust_remote_code=_is_qwen) model_name_or_path, trust_remote_code=_is_qwen)
if _is_qwen and self.tokenizer.pad_token is None: if _is_qwen and self.tokenizer.pad_token is None:
self.tokenizer.pad_token = self.tokenizer.eos_token self.tokenizer.pad_token = self.tokenizer.eos_token
# 自动判断架构：T5→T5EncoderModel；Qwen(decoder)→AutoModel(fp16)；其他→AutoModel
if _is_t5: if _is_t5:
self.encoder = T5EncoderModel.from_pretrained(model_name_or_path) self.encoder = T5EncoderModel.from_pretrained(model_name_or_path)
self.hidden_size = self.encoder.config.d_model self.hidden_size = self.encoder.config.d_model
elif _is_qwen and use_qlora:
from transformers import BitsAndBytesConfig
bnb = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_use_double_quant=True,
bnb_4bit_compute_dtype=torch.bfloat16,
)
self.encoder = AutoModel.from_pretrained(
model_name_or_path, trust_remote_code=True,
quantization_config=bnb, torch_dtype=torch.bfloat16)
self.hidden_size = self.encoder.config.hidden_size
elif _is_qwen: elif _is_qwen:
self.encoder = AutoModel.from_pretrained( self.encoder = AutoModel.from_pretrained(
model_name_or_path, trust_remote_code=True, torch_dtype=torch.float16) model_name_or_path, trust_remote_code=True, torch_dtype=torch.float16)
@ -61,31 +83,45 @@ class LLMPromptEncoder(nn.Module):
self.encoder = AutoModel.from_pretrained(model_name_or_path) self.encoder = AutoModel.from_pretrained(model_name_or_path)
self.hidden_size = self.encoder.config.hidden_size self.hidden_size = self.encoder.config.hidden_size
if use_lora: if self.use_lora:
self._apply_lora(lora_r, lora_alpha, lora_dropout) self._apply_lora(lora_r, lora_alpha, lora_dropout, prepare_kbit=use_qlora)
elif freeze: elif freeze:
for p in self.encoder.parameters(): for p in self.encoder.parameters():
p.requires_grad = False p.requires_grad = False
self._frozen = freeze and not use_lora self._frozen = freeze and not self.use_lora
# soft token 投影：chem'(d_model) / tab(d_model) -> hidden_size
if use_soft_prompt:
self.chem_to_llm = nn.Linear(d_model, self.hidden_size)
self.tab_to_llm = nn.Linear(d_model, self.hidden_size)
else:
self.chem_to_llm = None
self.tab_to_llm = None
self.proj_down = nn.Sequential( self.proj_down = nn.Sequential(
nn.Linear(self.hidden_size, d_model), nn.LayerNorm(d_model) nn.Linear(self.hidden_size, d_model), nn.LayerNorm(d_model)
) )
# 冻结特征缓存：smiles -> [H]（CPU）
# 缓存：冻结句向量缓存 / 文本 prompt 缓存
self._cache: Dict[str, torch.Tensor] = {} self._cache: Dict[str, torch.Tensor] = {}
# RAG 检索池（防泄漏：由 nested_cv 在每个 outer fold 用训练集设置） self._prompt_cache: Dict[str, str] = {}
# RAG 检索池
self._rag_pool_smiles: List[str] = [] self._rag_pool_smiles: List[str] = []
self._rag_pool_labels = None # np.ndarray [N] self._rag_pool_labels = None # delivery [N]
self._rag_pool_fps = None # 检索池指纹 self._rag_pool_extra = None # dict: task -> (values, valid)
self._rag_pool_id: str = "none" # 池子标识，用于缓存隔离 self._rag_pool_fps = None
self._rag_pool_id: str = "none"
def _apply_lora(self, r: int, alpha: int, dropout: float) -> None: def _apply_lora(self, r, alpha, dropout, prepare_kbit=False):
from peft import LoraConfig, get_peft_model from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
if prepare_kbit:
self.encoder = prepare_model_for_kbit_training(self.encoder)
else:
for p in self.encoder.parameters():
p.requires_grad = False
for p in self.encoder.parameters():
p.requires_grad = False
# 不同架构注意力层命名不同：
# T5: q/k/v/o；Roberta/ChemBERTa: query/key/value；Qwen/Llama: q_proj/k_proj/v_proj/o_proj
_enc_name = type(self.encoder).__name__.lower() _enc_name = type(self.encoder).__name__.lower()
if getattr(self, "_is_qwen", False) or "qwen" in _enc_name or "llama" in _enc_name: if getattr(self, "_is_qwen", False) or "qwen" in _enc_name or "llama" in _enc_name:
_targets = ["q_proj", "k_proj", "v_proj", "o_proj"] _targets = ["q_proj", "k_proj", "v_proj", "o_proj"]
@ -93,27 +129,27 @@ class LLMPromptEncoder(nn.Module):
_targets = ["q", "k", "v", "o"] _targets = ["q", "k", "v", "o"]
else: else:
_targets = ["query", "key", "value"] _targets = ["query", "key", "value"]
cfg = LoraConfig( cfg = LoraConfig(r=r, lora_alpha=alpha, lora_dropout=dropout,
r=r, lora_alpha=alpha, lora_dropout=dropout, target_modules=_targets, bias="none")
target_modules=_targets, bias="none",
)
self.encoder = get_peft_model(self.encoder, cfg) self.encoder = get_peft_model(self.encoder, cfg)
def set_retrieval_pool(self, smiles_list, labels, pool_id: str = "train") -> None: # ---------- 检索池 ----------
"""设置 RAG 检索池（防泄漏关键：只传训练集的 smiles 和 delivery 标签）。 def set_retrieval_pool(self, smiles_list, labels, pool_id="train", extra_labels=None):
labels: array-like [N] delivery 标签。pool_id: 池标识，切换时清 RAG 缓存。""" """设置 RAG 检索池（防泄漏：只传训练集分子与标签）。
labels: delivery [N]; extra_labels: dict task -> (values, valid_bool)。"""
import numpy as np import numpy as np
from lnp_ml.modeling.retrieval import _smiles_to_fp from lnp_ml.modeling.retrieval import _smiles_to_fp
self._rag_pool_smiles = list(smiles_list) self._rag_pool_smiles = list(smiles_list)
self._rag_pool_labels = np.asarray(labels, dtype=np.float32).reshape(-1) self._rag_pool_labels = np.asarray(labels, dtype=np.float32).reshape(-1)
self._rag_pool_extra = extra_labels
self._rag_pool_fps = [_smiles_to_fp(s) for s in self._rag_pool_smiles] self._rag_pool_fps = [_smiles_to_fp(s) for s in self._rag_pool_smiles]
if pool_id != self._rag_pool_id: if pool_id != self._rag_pool_id:
# 池子变了，清掉 RAG 编码缓存（避免跨 fold 串用）
self._cache = {k: v for k, v in self._cache.items() if not k.startswith("RAG::")} self._cache = {k: v for k, v in self._cache.items() if not k.startswith("RAG::")}
self._prompt_cache.clear()
self._rag_pool_id = pool_id self._rag_pool_id = pool_id
def _retrieve_topk(self, query_smiles: str): def _retrieve_topk(self, query_smiles: str):
"""检索 Top-K 相似邻居，返回 [(smiles, sim, label)]。排除查询分子自己（防泄漏）。""" """检索 Top-K 邻居，返回 [dict(smiles, sim, delivery, extra)]；排除查询分子自己。"""
from rdkit import DataStructs from rdkit import DataStructs
from lnp_ml.modeling.retrieval import _smiles_to_fp from lnp_ml.modeling.retrieval import _smiles_to_fp
qfp = _smiles_to_fp(query_smiles) qfp = _smiles_to_fp(query_smiles)
@ -121,105 +157,148 @@ class LLMPromptEncoder(nn.Module):
return [] return []
sims = [] sims = []
for j, (smi, fp) in enumerate(zip(self._rag_pool_smiles, self._rag_pool_fps)): for j, (smi, fp) in enumerate(zip(self._rag_pool_smiles, self._rag_pool_fps)):
if fp is None or smi == query_smiles: # 排除自己 if fp is None or smi == query_smiles:
continue continue
sims.append((j, DataStructs.TanimotoSimilarity(qfp, fp))) sims.append((j, DataStructs.TanimotoSimilarity(qfp, fp)))
sims.sort(key=lambda t: t[1], reverse=True) sims.sort(key=lambda t: t[1], reverse=True)
out = [] out = []
for j, sim in sims[:self.rag_top_k]: for j, sim in sims[: self.rag_top_k]:
out.append((self._rag_pool_smiles[j], sim, float(self._rag_pool_labels[j]))) extra = {}
if self._rag_pool_extra is not None:
for task in _EXTRA_TASKS:
if task in self._rag_pool_extra:
vals, valid = self._rag_pool_extra[task]
extra[task] = vals[j] if bool(valid[j]) else None
else:
extra[task] = None
out.append({
"smiles": self._rag_pool_smiles[j],
"sim": sim,
"delivery": float(self._rag_pool_labels[j]),
"extra": extra,
})
return out return out
@staticmethod
def _fmt(v, kind="float"):
if v is None:
return "unknown"
if kind == "int":
return str(int(v))
if kind == "vec":
return "[" + ", ".join(f"{x:.3f}" for x in v) + "]"
return f"{float(v):.3f}"
def _build_rag_prompt(self, target_smiles: str, neighbors) -> str: def _build_rag_prompt(self, target_smiles: str, neighbors) -> str:
"""构造 RAG prompt（与探针验证版一致，已验证 R²=0.1357）。""" """构造 RAG prompt：原始 SMILES + 邻居多任务结果（numeric）。"""
blocks = [] blocks = []
for rank, (smi, sim, lbl) in enumerate(neighbors, 1): for rank, nb in enumerate(neighbors, 1):
ex = nb["extra"]
blocks.append( blocks.append(
f"Retrieved sample {rank}:\nSMILES: {smi}\n" f"Retrieved sample {rank}:\n"
f"Similarity score: {sim:.3f}\nKnown outcome - delivery_log: {lbl:.3f}" f"SMILES: {nb['smiles']}\n"
f"Similarity score: {nb['sim']:.3f}\n"
f"delivery_log: {self._fmt(nb['delivery'])}\n"
f"size_z: {self._fmt(ex.get('size'))}\n"
f"pdi_class: {self._fmt(ex.get('pdi'), 'int')}\n"
f"ee_class: {self._fmt(ex.get('ee'), 'int')}\n"
f"toxic: {self._fmt(ex.get('toxic'), 'int')}\n"
f"biodist: {self._fmt(ex.get('biodist'), 'vec')}"
) )
retrieved_block = "\n\n".join(blocks) if blocks else "(no retrieved samples)" retrieved_block = "\n\n".join(blocks) if blocks else "(no retrieved samples)"
return ( return (
"Task: Encode the target LNP molecule into a retrieval-aware representation " "Task: Encode the target LNP molecule into a retrieval-aware representation "
"for downstream delivery prediction. Do not output predictions.\n\n" "for downstream multi-task property prediction. Do not output predictions.\n\n"
f"[Target Molecule]\nSMILES: {target_smiles}\n\n" f"[Target Molecule]\nSMILES: {target_smiles}\n\n"
"[Retrieved Similar LNP Samples]\n" "[Retrieved Similar LNP Samples]\n"
"The following are retrieved from the training set by fingerprint similarity, " "Retrieved from the training set by fingerprint similarity, with their known "
"with their known delivery outcomes:\n\n" "multi-task outcomes (delivery_log, size_z, pdi_class, ee_class, toxic, biodist; "
"'unknown' means the measurement is missing):\n\n"
f"{retrieved_block}\n\n" f"{retrieved_block}\n\n"
"[Encoding Instructions]\n" "[Encoding Instructions]\n"
"Capture the target molecular structure and the retrieval evidence " "Capture the target structure and the retrieval evidence (structural similarity "
"(structural similarity and consistency of retrieved delivery outcomes) " "and consistency of retrieved outcomes) into your internal representation."
"into your internal representation."
) )
@torch.no_grad() def _get_prompt(self, s: str) -> str:
def _encode_rag(self, smiles, device): if not self.use_rag:
"""RAG 编码：每个分子→检索→构造prompt→Qwen编码→取最后有效token。带缓存。""" return s
feats = [] key = f"{self._rag_pool_id}::{s}"
to_compute = [] if key not in self._prompt_cache:
keys = [] self._prompt_cache[key] = self._build_rag_prompt(s, self._retrieve_topk(s))
for s in smiles: return self._prompt_cache[key]
key = f"RAG::{self._rag_pool_id}::{s}"
keys.append(key)
if key not in self._cache:
to_compute.append(s)
# 逐个构造 prompt（检索是 CPU 操作）
if to_compute:
uniq = list(dict.fromkeys(to_compute))
prompts = [self._build_rag_prompt(s, self._retrieve_topk(s)) for s in uniq]
for i in range(0, len(uniq), 4):
bt = prompts[i:i+4]
enc = self.tokenizer(
bt, padding=True, truncation=True,
max_length=self.max_length, return_tensors="pt",
).to(device)
out = self.encoder(**enc).last_hidden_state # [B,L,H]
lengths = enc["attention_mask"].sum(1) - 1 # 最后有效token位置
b = torch.arange(out.size(0), device=device)
last_tok = out[b, lengths.long(), :] # [B,H]
for s, v in zip(uniq[i:i+4], last_tok):
self._cache[f"RAG::{self._rag_pool_id}::{s}"] = v.float().cpu()
return torch.stack([self._cache[k] for k in keys]).to(device)
def _mean_pool(self, last_hidden: torch.Tensor, mask: torch.Tensor) -> torch.Tensor: # ---------- soft-prompt 编码（带梯度，不缓存特征）----------
m = mask.unsqueeze(-1).float() # [B, L, 1] def _encode_softrag(self, smiles, chem, tab, device) -> torch.Tensor:
prompts = [self._get_prompt(s) for s in smiles]
enc = self.tokenizer(
prompts, padding=True, truncation=True,
max_length=self.max_length, return_tensors="pt",
).to(device)
embed_layer = self.encoder.get_input_embeddings()
text_embeds = embed_layer(enc["input_ids"]) # [B, L, H]
text_mask = enc["attention_mask"] # [B, L]
soft_list = []
if self.chem_to_llm is not None and chem is not None:
soft_list.append(self.chem_to_llm(chem)) # [B, n_chem, H]
if self.tab_to_llm is not None and tab is not None:
soft_list.append(self.tab_to_llm(tab)) # [B, n_tab, H]
if soft_list:
soft = torch.cat(soft_list, dim=1).to(text_embeds.dtype)
inputs_embeds = torch.cat([soft, text_embeds], dim=1)
soft_mask = torch.ones(soft.size(0), soft.size(1),
device=device, dtype=text_mask.dtype)
attn_mask = torch.cat([soft_mask, text_mask], dim=1)
else:
inputs_embeds = text_embeds
attn_mask = text_mask
out = self.encoder(inputs_embeds=inputs_embeds, attention_mask=attn_mask).last_hidden_state
lengths = attn_mask.sum(1) - 1
b = torch.arange(out.size(0), device=device)
feat = out[b, lengths.long(), :] # [B, H] 最后有效 token
return self.proj_down(feat.float())
# ---------- 旧路径（保留，向后兼容）----------
def _mean_pool(self, last_hidden, mask):
m = mask.unsqueeze(-1).float()
return (last_hidden * m).sum(1) / m.sum(1).clamp(min=1e-6) return (last_hidden * m).sum(1) / m.sum(1).clamp(min=1e-6)
@torch.no_grad() @torch.no_grad()
def _encode_frozen(self, smiles: List[str], device: torch.device) -> torch.Tensor: def _encode_frozen(self, smiles, device):
missing = [s for s in smiles if s not in self._cache] missing = [s for s in smiles if s not in self._cache]
if missing: if missing:
uniq = list(dict.fromkeys(missing)) uniq = list(dict.fromkeys(missing))
for i in range(0, len(uniq), 256): for i in range(0, len(uniq), 256):
chunk = uniq[i:i + 256] chunk = uniq[i:i + 256]
enc = self.tokenizer( enc = self.tokenizer(chunk, padding=True, truncation=True,
chunk, padding=True, truncation=True, max_length=self.max_length, return_tensors="pt").to(device)
max_length=self.max_length, return_tensors="pt",
).to(device)
out = self.encoder(**enc).last_hidden_state out = self.encoder(**enc).last_hidden_state
pooled = self._mean_pool(out, enc["attention_mask"]) pooled = self._mean_pool(out, enc["attention_mask"])
for s, v in zip(chunk, pooled): for s, v in zip(chunk, pooled):
self._cache[s] = v.cpu() self._cache[s] = v.cpu()
return torch.stack([self._cache[s] for s in smiles]).to(device) return torch.stack([self._cache[s] for s in smiles]).to(device)
def _encode_trainable(self, smiles: List[str], device: torch.device) -> torch.Tensor: def _encode_trainable(self, smiles, device):
enc = self.tokenizer( enc = self.tokenizer(list(smiles), padding=True, truncation=True,
list(smiles), padding=True, truncation=True, max_length=self.max_length, return_tensors="pt").to(device)
max_length=self.max_length, return_tensors="pt",
).to(device)
out = self.encoder(**enc).last_hidden_state out = self.encoder(**enc).last_hidden_state
return self._mean_pool(out, enc["attention_mask"]) return self._mean_pool(out, enc["attention_mask"])
def forward(self, smiles: List[str], tab: Optional[torch.Tensor] = None) -> torch.Tensor: def forward(self, smiles: List[str], chem: Optional[torch.Tensor] = None,
"""Args: smiles [B] SMILES 字符串列表。Returns: [B, d_model]。""" tab: Optional[torch.Tensor] = None) -> torch.Tensor:
device = self.proj_down[0].weight.device device = self.proj_down[0].weight.device
if self.use_soft_prompt:
return self._encode_softrag(smiles, chem, tab, device)
if self.use_rag: if self.use_rag:
feat = self._encode_rag(smiles, device) return self._encode_softrag(smiles, None, None, device)
else: feat = self._encode_frozen(smiles, device) if self._frozen \
feat = self._encode_frozen(smiles, device) if self._frozen \ else self._encode_trainable(smiles, device)
else self._encode_trainable(smiles, device)
return self.proj_down(feat) return self.proj_down(feat)
def clear_cache(self) -> None: def clear_cache(self) -> None:
self._cache.clear() self._cache.clear()
self._prompt_cache.clear()

38
lnp_ml/modeling/models.py
View File

@ -20,17 +20,26 @@ PoolingStrategy = Literal["attention", "avg", "max"]
# Token 维度配置 # Token 维度配置
def _infer_desc_dim() -> int:
"""RDKit 描述符数量随版本变化，运行时实测，避免写死导致 BatchNorm 维度不匹配。"""
from rdkit import Chem
from rdkit.Chem import Descriptors
return len(Descriptors.CalcMolDescriptors(Chem.MolFromSmiles("CCO")))
_DESC_DIM = _infer_desc_dim() # 本机 = 210
DEFAULT_INPUT_DIMS = { DEFAULT_INPUT_DIMS = {
# Channel A: 化学特征 # Channel A: 化学特征
"mpnn": 600, # D-MPNN embedding "mpnn": 600, # D-MPNN embedding
"morgan": 1024, # Morgan fingerprint "morgan": 1024, # Morgan fingerprint
"maccs": 167, # MACCS keys "maccs": 167, # MACCS keys
"desc": 217, # RDKit descriptors "desc": _DESC_DIM, # RDKit descriptors（随版本动态）
# Channel B: 配方/实验条件 # Channel B: 配方/实验条件
"comp": 5, # 配方比例 "comp": 5, # 配方比例
"phys": 12, # 物理参数 one-hot "phys": 12, # 物理参数 one-hot
"help": 4, # Helper lipid one-hot "help": 4, # Helper lipid one-hot
"exp": 32, # 实验条件 one-hot "exp": 32, # 实验条件 one-hot
} }
# 化学 / 配方 token 的键顺序 # 化学 / 配方 token 的键顺序
@ -95,6 +104,8 @@ class LNPModel(nn.Module):
llm_model_path: str = DEFAULT_MOLT5_PATH, llm_model_path: str = DEFAULT_MOLT5_PATH,
llm_freeze: bool = True, llm_freeze: bool = True,
llm_use_lora: bool = False, llm_use_lora: bool = False,
llm_use_qlora: bool = False,
use_soft_prompt: bool = False,
llm_lora_r: int = 8, llm_lora_r: int = 8,
llm_lora_alpha: int = 16, llm_lora_alpha: int = 16,
llm_lora_dropout: float = 0.05, llm_lora_dropout: float = 0.05,
@ -171,11 +182,13 @@ class LNPModel(nn.Module):
model_name_or_path=llm_model_path, model_name_or_path=llm_model_path,
freeze=llm_freeze, freeze=llm_freeze,
use_lora=llm_use_lora, use_lora=llm_use_lora,
use_qlora=llm_use_qlora,
lora_r=llm_lora_r, lora_r=llm_lora_r,
lora_alpha=llm_lora_alpha, lora_alpha=llm_lora_alpha,
lora_dropout=llm_lora_dropout, lora_dropout=llm_lora_dropout,
use_rag=use_rag, use_rag=use_rag,
rag_top_k=rag_top_k, rag_top_k=rag_top_k,
use_soft_prompt=use_soft_prompt,
) )
else: else:
self.llm_prompt = None self.llm_prompt = None
@ -246,7 +259,10 @@ class LNPModel(nn.Module):
f_llm = None f_llm = None
if self.llm_prompt is not None and smiles is not None: if self.llm_prompt is not None and smiles is not None:
f_llm = self.llm_prompt(smiles) if getattr(self.llm_prompt, "use_soft_prompt", False):
f_llm = self.llm_prompt(smiles, chem=chem, tab=tab)
else:
f_llm = self.llm_prompt(smiles)
f_retr = None f_retr = None
if self.retr_proj is not None and self.retriever is not None and smiles is not None: if self.retr_proj is not None and self.retriever is not None and smiles is not None:
@ -432,6 +448,8 @@ class LNPModelWithoutMPNN(LNPModel):
llm_model_path: str = DEFAULT_MOLT5_PATH, llm_model_path: str = DEFAULT_MOLT5_PATH,
llm_freeze: bool = True, llm_freeze: bool = True,
llm_use_lora: bool = False, llm_use_lora: bool = False,
llm_use_qlora: bool = False,
use_soft_prompt: bool = False,
llm_lora_r: int = 8, llm_lora_r: int = 8,
llm_lora_alpha: int = 16, llm_lora_alpha: int = 16,
llm_lora_dropout: float = 0.05, llm_lora_dropout: float = 0.05,
@ -467,6 +485,8 @@ class LNPModelWithoutMPNN(LNPModel):
llm_model_path=llm_model_path, llm_model_path=llm_model_path,
llm_freeze=llm_freeze, llm_freeze=llm_freeze,
llm_use_lora=llm_use_lora, llm_use_lora=llm_use_lora,
llm_use_qlora=llm_use_qlora,
use_soft_prompt=use_soft_prompt,
llm_lora_r=llm_lora_r, llm_lora_r=llm_lora_r,
llm_lora_alpha=llm_lora_alpha, llm_lora_alpha=llm_lora_alpha,
llm_lora_dropout=llm_lora_dropout, llm_lora_dropout=llm_lora_dropout,

82
lnp_ml/modeling/nested_cv_optuna.py
View File

@ -246,6 +246,26 @@ def create_model(
) )
def _build_rag_pool(full_dataset, idx):
"""从 dataset + 索引构造 RAG 池：返回 (smiles, delivery[N], extra_labels)。"""
import numpy as np
idx = np.asarray(idx)
smiles = [full_dataset.smiles[i] for i in idx]
delivery = full_dataset.delivery[idx].reshape(-1)
extra = {}
if full_dataset.size is not None:
extra["size"] = (full_dataset.size[idx], ~np.isnan(full_dataset.size[idx]))
if full_dataset.pdi is not None:
extra["pdi"] = (full_dataset.pdi[idx], full_dataset.pdi_valid[idx])
if full_dataset.ee is not None:
extra["ee"] = (full_dataset.ee[idx], full_dataset.ee_valid[idx])
if full_dataset.toxic is not None:
extra["toxic"] = (full_dataset.toxic[idx], full_dataset.toxic[idx] >= 0)
if full_dataset.biodist is not None:
extra["biodist"] = (full_dataset.biodist[idx], full_dataset.biodist_valid[idx])
return smiles, delivery, extra
# ============ 评估指标 ============ # ============ 评估指标 ============
def evaluate_on_test( def evaluate_on_test(
@ -496,7 +516,8 @@ def run_inner_optuna(
val_subset = Subset(full_dataset, actual_val_idx.tolist()) val_subset = Subset(full_dataset, actual_val_idx.tolist())
train_loader = DataLoader( train_loader = DataLoader(
train_subset, batch_size=batch_size, shuffle=True, collate_fn=collate_fn train_subset, batch_size=batch_size, shuffle=True,
collate_fn=collate_fn, drop_last=True,
) )
val_loader = DataLoader( val_loader = DataLoader(
val_subset, batch_size=batch_size, shuffle=False, collate_fn=collate_fn val_subset, batch_size=batch_size, shuffle=False, collate_fn=collate_fn
@ -525,6 +546,12 @@ def run_inner_optuna(
) )
if rdkit_cache is not None: if rdkit_cache is not None:
model.rdkit_encoder._cache = rdkit_cache model.rdkit_encoder._cache = rdkit_cache
# 内层 RAG 池：只用内层训练集（防 inner-val 泄漏）
if llm_kwargs_t.get("use_rag", False) and getattr(model, "llm_prompt", None) is not None:
_s, _d, _ex = _build_rag_pool(full_dataset, actual_train_idx)
model.llm_prompt.set_retrieval_pool(
_s, _d, pool_id=f"inner{inner_fold}", extra_labels=_ex)
# 加载预训练权重 # 加载预训练权重
if pretrain_state_dict is not None and pretrain_config is not None: if pretrain_state_dict is not None and pretrain_config is not None:
@ -631,6 +658,29 @@ def _run_single_outer_fold(
fold_dir = Path(fold_dir) fold_dir = Path(fold_dir)
fold_dir.mkdir(parents=True, exist_ok=True) fold_dir.mkdir(parents=True, exist_ok=True)
# ===== 断点续跑 =====
_f_metrics = fold_dir / "test_metrics.json"
_f_bp = fold_dir / "best_params.json"
_f_em = fold_dir / "epoch_mean.json"
# (1) 整折已完成 → 读回结果直接跳过
if _f_metrics.exists() and _f_bp.exists() and _f_em.exists():
logger.info(f"[RESUME] Outer fold {outer_fold} 已完成，跳过。")
with open(_f_metrics) as f:
_tm = json.load(f)
with open(_f_bp) as f:
_bp = json.load(f)
with open(_f_em) as f:
_em = json.load(f)["epoch_mean"]
return {"fold": outer_fold, "best_params": _bp,
"epoch_mean": _em, "test_metrics": _tm}
# (2) 内层已完成、外层中断 → 复用超参，跳过内层 Optuna
if precomputed_best_params is None and _f_bp.exists() and _f_em.exists():
with open(_f_bp) as f:
precomputed_best_params = json.load(f)
with open(_f_em) as f:
precomputed_epoch_mean = json.load(f)["epoch_mean"]
logger.info(f"[RESUME] fold {outer_fold} 复用已存 best_params，跳过内层 Optuna。")
full_dataset = LNPDataset(df) full_dataset = LNPDataset(df)
logger.info(f"\n{'='*60}") logger.info(f"\n{'='*60}")
@ -703,7 +753,8 @@ def _run_single_outer_fold(
test_subset = Subset(full_dataset, outer_test_idx.tolist()) test_subset = Subset(full_dataset, outer_test_idx.tolist())
train_loader = DataLoader( train_loader = DataLoader(
train_subset, batch_size=batch_size, shuffle=True, collate_fn=collate_fn train_subset, batch_size=batch_size, shuffle=True,
collate_fn=collate_fn, drop_last=True,
) )
monitor_loader = DataLoader( monitor_loader = DataLoader(
monitor_subset, batch_size=batch_size, shuffle=False, collate_fn=collate_fn monitor_subset, batch_size=batch_size, shuffle=False, collate_fn=collate_fn
@ -748,11 +799,10 @@ def _run_single_outer_fold(
# ===== RAG 检索池设置（防泄漏：只用训练集分子+标签）===== # ===== RAG 检索池设置（防泄漏：只用训练集分子+标签）=====
if llm_kwargs and llm_kwargs.get("use_rag", False) and getattr(model, "llm_prompt", None) is not None: if llm_kwargs and llm_kwargs.get("use_rag", False) and getattr(model, "llm_prompt", None) is not None:
_rag_smiles = [full_dataset.smiles[i] for i in outer_train_idx] _s, _d, _ex = _build_rag_pool(full_dataset, outer_train_idx)
_rag_labels = full_dataset.delivery[outer_train_idx].reshape(-1)
model.llm_prompt.set_retrieval_pool( model.llm_prompt.set_retrieval_pool(
_rag_smiles, _rag_labels, pool_id=f"fold{outer_fold}") _s, _d, pool_id=f"fold{outer_fold}", extra_labels=_ex)
logger.info(f"[RAG] 检索池={len(_rag_smiles)} 训练分子（不含测试集，防泄漏）") logger.info(f"[RAG] 检索池={len(_s)} 训练分子（多任务标签，不含测试集，防泄漏）")
if pretrain_state_dict is not None and pretrain_config is not None: if pretrain_state_dict is not None and pretrain_config is not None:
loaded = load_pretrain_weights_to_model( loaded = load_pretrain_weights_to_model(
@ -776,7 +826,7 @@ def _run_single_outer_fold(
freeze_backbone_epochs=3, freeze_backbone_epochs=3,
) )
model.load_state_dict(train_result["final_state"]) model.load_state_dict(train_result["final_state"], strict=False)
model = model.to(device) model = model.to(device)
config = { config = {
@ -838,6 +888,7 @@ def _run_single_outer_fold(
def main( def main(
input_path: Path = INTERIM_DATA_DIR / "internal.csv", input_path: Path = INTERIM_DATA_DIR / "internal.csv",
output_dir: Path = MODELS_DIR / "nested_cv", output_dir: Path = MODELS_DIR / "nested_cv",
resume_dir: Optional[Path] = None,
# CV 参数 # CV 参数
n_outer_folds: int = 5, n_outer_folds: int = 5,
n_inner_folds: int = 3, n_inner_folds: int = 3,
@ -873,6 +924,8 @@ def main(
llm_model_path: str = DEFAULT_MOLT5_PATH, llm_model_path: str = DEFAULT_MOLT5_PATH,
llm_freeze: bool = True, llm_freeze: bool = True,
llm_use_lora: bool = False, llm_use_lora: bool = False,
llm_use_qlora: bool = False,
use_soft_prompt: bool = False,
llm_lora_r: int = 8, llm_lora_r: int = 8,
llm_lora_alpha: int = 16, llm_lora_alpha: int = 16,
llm_lora_dropout: float = 0.05, llm_lora_dropout: float = 0.05,
@ -905,6 +958,8 @@ def main(
llm_model_path=llm_model_path, llm_model_path=llm_model_path,
llm_freeze=llm_freeze, llm_freeze=llm_freeze,
llm_use_lora=llm_use_lora, llm_use_lora=llm_use_lora,
llm_use_qlora=llm_use_qlora,
use_soft_prompt=use_soft_prompt,
llm_lora_r=llm_lora_r, llm_lora_r=llm_lora_r,
llm_lora_alpha=llm_lora_alpha, llm_lora_alpha=llm_lora_alpha,
llm_lora_dropout=llm_lora_dropout, llm_lora_dropout=llm_lora_dropout,
@ -923,10 +978,15 @@ def main(
else: else:
logger.warning(f"Pretrain checkpoint not found: {init_from_pretrain}, skipping") logger.warning(f"Pretrain checkpoint not found: {init_from_pretrain}, skipping")
# 创建输出目录（带时间戳） # 创建输出目录（带时间戳；--resume-dir 指定则复用，支持断点续跑）
run_name = datetime.now().strftime("%Y%m%d_%H%M%S") if resume_dir is not None:
run_dir = output_dir / run_name run_dir = Path(resume_dir)
run_dir.mkdir(parents=True, exist_ok=True) run_dir.mkdir(parents=True, exist_ok=True)
logger.info(f"[RESUME] 复用已有运行目录: {run_dir}")
else:
run_name = datetime.now().strftime("%Y%m%d_%H%M%S")
run_dir = output_dir / run_name
run_dir.mkdir(parents=True, exist_ok=True)
logger.info(f"Output directory: {run_dir}") logger.info(f"Output directory: {run_dir}")
# 加载数据 # 加载数据

85
lnp_ml/modeling/retrieval.py Normal file
View File

@ -0,0 +1,85 @@
"""Morgan 指纹检索：RAG prompt 构造与检索旁路共用。
- _smiles_to_fp: SMILES -> RDKit ExplicitBitVect（供 Tanimoto 相似度）。
- MorganRetriever: 训练集指纹检索器，query_batch 返回每个查询的 3 维检索特征。
"""
from typing import List, Optional
import numpy as np
from rdkit import Chem, DataStructs
from rdkit.Chem import AllChem
# 与 RDKitFeaturizer 的 morgan token 保持一致
MORGAN_RADIUS = 2
MORGAN_NBITS = 1024
def _smiles_to_fp(smiles: Optional[str], radius: int = MORGAN_RADIUS, n_bits: int = MORGAN_NBITS):
"""SMILES -> Morgan ExplicitBitVect；非法/空返回 None。"""
if not smiles:
return None
mol = Chem.MolFromSmiles(smiles)
if mol is None:
return None
return AllChem.GetMorganFingerprintAsBitVect(mol, radius=radius, nBits=n_bits)
class MorganRetriever:
"""基于 Morgan + Tanimoto 的训练集检索器（用于 use_retrieval 旁路）。
query_batch 为每个查询返回 3 维特征：
[相似度加权邻居标签均值, top1 相似度, top-k 平均相似度]
"""
def __init__(self, smiles_list: List[str], labels, k: int = 5) -> None:
self.k = int(k)
self.pool_smiles = list(smiles_list)
labels = np.asarray(labels, dtype=np.float32)
if labels.ndim == 1:
labels = labels.reshape(-1, 1)
self.pool_labels = labels # [N, D]，旁路只用第 0 列
fps = [_smiles_to_fp(s) for s in self.pool_smiles]
# 只保留可解析的池分子
self._valid_pos = [i for i, fp in enumerate(fps) if fp is not None]
self._valid_fps = [fps[i] for i in self._valid_pos]
self._valid_labels = self.pool_labels[self._valid_pos, 0] if self._valid_pos else np.zeros(0, np.float32)
self._valid_smiles = [self.pool_smiles[i] for i in self._valid_pos]
self._fp_cache = {} # query 指纹缓存
def _query_fp(self, smiles: str):
if smiles not in self._fp_cache:
self._fp_cache[smiles] = _smiles_to_fp(smiles)
return self._fp_cache[smiles]
def query_batch(self, smiles_list: List[str], exclude_self="auto") -> np.ndarray:
return np.stack([self._query_one(s, exclude_self) for s in smiles_list]).astype(np.float32)
def _query_one(self, smiles: str, exclude_self) -> np.ndarray:
qfp = self._query_fp(smiles)
if qfp is None or not self._valid_fps:
return np.zeros(3, dtype=np.float32)
# 向量化 Tanimoto（C 层批量），替代 Python 逐条
sims = np.asarray(DataStructs.BulkTanimotoSimilarity(qfp, self._valid_fps), dtype=np.float32)
if exclude_self:
self_mask = np.fromiter((s == smiles for s in self._valid_smiles), dtype=bool, count=len(sims))
sims = np.where(self_mask, -1.0, sims)
k = min(self.k, int((sims >= 0).sum()))
if k <= 0:
return np.zeros(3, dtype=np.float32)
# argpartition 取 top-k（O(N)），再对这 k 个排序
top = np.argpartition(-sims, k - 1)[:k]
top = top[np.argsort(-sims[top])]
top_sims = sims[top]
top_labels = self._valid_labels[top]
w = np.clip(top_sims, 0.0, None)
wsum = float(w.sum())
weighted_label = float((w * top_labels).sum() / wsum) if wsum > 0 else float(top_labels.mean())
return np.array([weighted_label, float(top_sims[0]), float(top_sims.mean())], dtype=np.float32)

6
lnp_ml/modeling/trainer_balanced.py
View File

@ -406,7 +406,7 @@ def train_with_early_stopping(
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau( scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
optimizer, mode="min", factor=0.5, patience=5 optimizer, mode="min", factor=0.5, patience=5
) )
early_stopping = EarlyStoppingBalanced(patience=patience) early_stopping = EarlyStoppingBalanced(patience=patience, min_delta=1e-3)
history = {"train": [], "val": []} history = {"train": [], "val": []}
best_val_loss = float("inf") best_val_loss = float("inf")
@ -438,9 +438,9 @@ def train_with_early_stopping(
if early_stopping(val_metrics["loss"], epoch): if early_stopping(val_metrics["loss"], epoch):
break break
# Restore best model # Restore best model（QLoRA 4-bit 基座的量化元数据键用 strict=False 忽略）
if best_state is not None: if best_state is not None:
model.load_state_dict(best_state) model.load_state_dict(best_state, strict=False)
return { return {
"history": history, "history": history,

12
models/abl/baseline/20260617_180556/outer_fold_0/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.28242799368681437,
"lr": 0.00037183641805732076,
"weight_decay": 6.290644294586152e-05,
"backbone_lr_ratio": 0.10677482709481352,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

42
models/abl/baseline/20260617_180556/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.32790454280151365,
"rmse": 0.572629498717551,
"mae": 0.35871042688208893,
"r2": -0.08876073797082196
},
"delivery": {
"n_samples": 58,
"mse": 0.7250728333301815,
"rmse": 0.8515120864263651,
"mae": 0.7098816664696768,
"r2": 0.07838867510225656
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.3435971685971686,
"recall": 0.5359094457455114,
"f1": 0.34977324263038545
},
"ee": {
"n_samples": 84,
"accuracy": 0.7142857142857143,
"precision": 0.6871101871101871,
"recall": 0.707142857142857,
"f1": 0.6756066411238826
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.4073394583325288,
"js_divergence": 0.092887269703335
}
}

12
models/abl/baseline/20260617_180556/outer_fold_1/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.23085562232445592,
"lr": 0.0005227270589225132,
"weight_decay": 0.00461701039547356,
"backbone_lr_ratio": 0.12087164274488635,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

42
models/abl/baseline/20260617_180556/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.233876145855699,
"rmse": 0.4836074294876982,
"mae": 0.3760785318556286,
"r2": -2.874543407510119
},
"delivery": {
"n_samples": 61,
"mse": 1.0238964615450643,
"rmse": 1.0118776910007772,
"mae": 0.6578027546466862,
"r2": 0.2249841902125319
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5238095238095238,
"precision": 0.34497354497354493,
"recall": 0.3280423280423281,
"f1": 0.3231922398589065
},
"ee": {
"n_samples": 84,
"accuracy": 0.5595238095238095,
"precision": 0.5101731601731602,
"recall": 0.5036154321868608,
"f1": 0.48790324563862697
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.48528378254405913,
"js_divergence": 0.12258377401744426
}
}

12
models/abl/baseline/20260617_180556/outer_fold_2/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.18117503720006686,
"lr": 8.135394049241399e-05,
"weight_decay": 0.015076469897649124,
"backbone_lr_ratio": 0.9586175198423679,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

42
models/abl/baseline/20260617_180556/outer_fold_2/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.07870736214858826,
"rmse": 0.28054832408800495,
"mae": 0.2252055633635748,
"r2": 0.15667568332606574
},
"delivery": {
"n_samples": 60,
"mse": 0.6624288051399898,
"rmse": 0.8138972939750997,
"mae": 0.5578871982870623,
"r2": 0.118192445613993
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5714285714285714,
"precision": 0.43453195231581004,
"recall": 0.7014848950332823,
"f1": 0.4452380952380952
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6140109890109889,
"recall": 0.6785714285714285,
"f1": 0.6262626262626263
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3967825471495568,
"js_divergence": 0.1057568924841401
}
}

12
models/abl/baseline/20260617_180556/outer_fold_3/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.1799936812783694,
"lr": 0.00033865324369584493,
"weight_decay": 0.01440133094015265,
"backbone_lr_ratio": 0.3278334190266755,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

42
models/abl/baseline/20260617_180556/outer_fold_3/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.477705869562307,
"rmse": 0.6911626939891266,
"mae": 0.38659381004701177,
"r2": -0.46585807501413723
},
"delivery": {
"n_samples": 59,
"mse": 0.7692103143544243,
"rmse": 0.8770463581558413,
"mae": 0.6155587352528158,
"r2": 0.2522361838732067
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.4426767676767677,
"recall": 0.6817716333845366,
"f1": 0.43613724611708476
},
"ee": {
"n_samples": 84,
"accuracy": 0.6428571428571429,
"precision": 0.5650987224157956,
"recall": 0.5790095525389644,
"f1": 0.5637508626639062
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.4102080974965646,
"js_divergence": 0.11100613154873733
}
}

12
models/abl/baseline/20260617_180556/outer_fold_4/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.14162975892531532,
"lr": 0.0004019907813153378,
"weight_decay": 0.00037960139235180325,
"backbone_lr_ratio": 0.18010931581670617,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

42
models/abl/baseline/20260617_180556/outer_fold_4/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.1842198466764517,
"rmse": 0.4292083953937198,
"mae": 0.29178874833243235,
"r2": -0.33736192476816806
},
"delivery": {
"n_samples": 58,
"mse": 0.9003281069383875,
"rmse": 0.9488562098328637,
"mae": 0.6070586243837044,
"r2": 0.1293975832987566
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.34548450889914306,
"recall": 0.5337941628264209,
"f1": 0.3497882046269143
},
"ee": {
"n_samples": 84,
"accuracy": 0.7023809523809523,
"precision": 0.6548765244417418,
"recall": 0.7576923076923077,
"f1": 0.6777508856852221
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3946735882065513,
"js_divergence": 0.1067245125767987
}
}

12
models/abl/baseline/20260617_180556/repeat_1/outer_fold_0/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.28242799368681437,
"lr": 0.00037183641805732076,
"weight_decay": 6.290644294586152e-05,
"backbone_lr_ratio": 0.10677482709481352,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_1/outer_fold_0/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72, 73, 75, 76, 77, 80, 81, 85, 86, 88, 89, 90, 92, 94, 95, 96, 98, 99, 100, 101, 103, 104, 105, 106, 107, 108, 109, 111, 113, 114, 115, 116, 117, 118, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 133, 134, 135, 136, 138, 139, 141, 144, 145, 146, 147, 148, 149, 150, 151, 153, 154, 155, 156, 157, 158, 159, 160, 163, 164, 167, 168, 169, 170, 171, 172, 175, 176, 178, 179, 180, 182, 183, 184, 185, 188, 189, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 206, 208, 210, 211, 212, 213, 216, 217, 218, 219, 220, 222, 224, 225, 227, 228, 229, 230, 231, 232, 233, 234, 236, 237, 239, 241, 242, 244, 245, 247, 248, 249, 250, 252, 253, 255, 256, 257, 258, 259, 260, 261, 262, 265, 267, 268, 269, 270, 271, 272, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 285, 286, 287, 288, 291, 292, 294, 296, 297, 298, 299, 300, 301, 303, 304, 306, 307, 308, 309, 310, 311, 312, 313, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 333, 335, 336, 337, 338, 339, 340, 342, 343, 344, 345, 346, 347, 348, 349, 351, 352, 354, 355, 356, 357, 358, 359, 360, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372, 374, 375, 377, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 393, 395, 396, 397, 398, 399, 400, 401, 403, 404, 405, 406, 407, 408, 409, 411, 412, 413, 414, 415, 417, 418, 419], "outer_test_idx": [7, 16, 20, 29, 51, 59, 60, 69, 70, 74, 78, 79, 82, 83, 84, 87, 91, 93, 97, 102, 110, 112, 119, 120, 121, 132, 137, 140, 142, 143, 152, 161, 162, 165, 166, 173, 174, 177, 181, 186, 187, 194, 205, 207, 209, 214, 215, 221, 223, 226, 235, 238, 240, 243, 246, 251, 254, 263, 264, 266, 273, 284, 289, 290, 293, 295, 302, 305, 314, 315, 331, 334, 341, 350, 353, 361, 373, 376, 378, 392, 394, 402, 410, 416]}

42
models/abl/baseline/20260617_180556/repeat_1/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.3808128541133737,
"rmse": 0.6171003598389598,
"mae": 0.4168719153806388,
"r2": -0.2644353156285024
},
"delivery": {
"n_samples": 58,
"mse": 0.751610048614176,
"rmse": 0.8669544674400012,
"mae": 0.7122741444999802,
"r2": 0.04465827311677528
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5,
"precision": 0.3181788931788932,
"recall": 0.5039032006245121,
"f1": 0.32369614512471656
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6258687258687259,
"recall": 0.6461904761904762,
"f1": 0.6127803231251508
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.39644445917049376,
"js_divergence": 0.09334798400608228
}
}

12
models/abl/baseline/20260617_180556/repeat_1/outer_fold_1/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.23085562232445592,
"lr": 0.0005227270589225132,
"weight_decay": 0.00461701039547356,
"backbone_lr_ratio": 0.12087164274488635,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_1/outer_fold_1/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 3, 5, 6, 7, 10, 11, 12, 13, 15, 16, 18, 19, 20, 21, 24, 26, 27, 29, 30, 31, 32, 33, 35, 36, 38, 39, 40, 42, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 97, 98, 99, 100, 101, 102, 103, 105, 106, 107, 108, 109, 110, 112, 114, 115, 118, 119, 120, 121, 123, 124, 125, 126, 128, 129, 130, 131, 132, 136, 137, 140, 141, 142, 143, 144, 145, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 161, 162, 164, 165, 166, 167, 168, 169, 171, 172, 173, 174, 175, 176, 177, 178, 179, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 194, 195, 196, 197, 198, 199, 201, 202, 204, 205, 206, 207, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 244, 246, 247, 248, 250, 251, 252, 254, 255, 256, 258, 259, 260, 261, 262, 263, 264, 266, 268, 269, 270, 271, 273, 274, 275, 276, 277, 278, 279, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 300, 301, 302, 303, 304, 305, 306, 307, 308, 310, 311, 313, 314, 315, 316, 317, 318, 319, 321, 322, 323, 324, 325, 327, 328, 329, 330, 331, 333, 334, 335, 337, 338, 339, 340, 341, 342, 343, 344, 345, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 361, 362, 364, 365, 366, 367, 368, 369, 373, 374, 375, 376, 378, 379, 380, 381, 382, 384, 387, 388, 389, 391, 392, 393, 394, 396, 397, 398, 399, 402, 403, 404, 405, 406, 408, 409, 410, 411, 412, 413, 414, 416, 417, 419], "outer_test_idx": [1, 2, 4, 8, 9, 14, 17, 22, 23, 25, 28, 34, 37, 41, 43, 45, 63, 64, 68, 85, 96, 104, 111, 113, 116, 117, 122, 127, 133, 134, 135, 138, 139, 147, 160, 163, 170, 180, 191, 192, 193, 200, 203, 208, 224, 227, 230, 239, 241, 245, 249, 253, 257, 265, 267, 272, 280, 294, 298, 299, 309, 312, 320, 326, 332, 336, 346, 347, 360, 363, 370, 371, 372, 377, 383, 385, 386, 390, 395, 400, 401, 407, 415, 418]}

42
models/abl/baseline/20260617_180556/repeat_1/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.3074589292084139,
"rmse": 0.5544897917982025,
"mae": 0.3793355481965201,
"r2": -4.093563359726248
},
"delivery": {
"n_samples": 61,
"mse": 1.0592195690639976,
"rmse": 1.029183933543464,
"mae": 0.6786946221025753,
"r2": 0.19824714422578682
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5952380952380952,
"precision": 0.37142213207121927,
"recall": 0.35978835978835977,
"f1": 0.35976608187134507
},
"ee": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.5052287581699346,
"recall": 0.5094429380143666,
"f1": 0.4903811048389362
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.5219711265981701,
"js_divergence": 0.13035793372268764
}
}

12
models/abl/baseline/20260617_180556/repeat_1/outer_fold_2/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.18117503720006686,
"lr": 8.135394049241399e-05,
"weight_decay": 0.015076469897649124,
"backbone_lr_ratio": 0.9586175198423679,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_1/outer_fold_2/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 36, 37, 38, 39, 40, 41, 43, 44, 45, 47, 48, 49, 50, 51, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 68, 69, 70, 72, 74, 76, 78, 79, 80, 82, 83, 84, 85, 86, 87, 88, 91, 93, 94, 95, 96, 97, 100, 101, 102, 103, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 142, 143, 145, 147, 148, 149, 150, 151, 152, 153, 156, 157, 158, 159, 160, 161, 162, 163, 165, 166, 167, 168, 170, 171, 172, 173, 174, 175, 177, 179, 180, 181, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 196, 197, 199, 200, 202, 203, 204, 205, 207, 208, 209, 210, 211, 212, 213, 214, 215, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 232, 233, 235, 236, 237, 238, 239, 240, 241, 242, 243, 245, 246, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 264, 265, 266, 267, 268, 269, 271, 272, 273, 274, 275, 276, 280, 281, 282, 283, 284, 285, 287, 289, 290, 293, 294, 295, 296, 297, 298, 299, 300, 302, 303, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 317, 318, 320, 321, 322, 323, 324, 326, 327, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 358, 360, 361, 363, 364, 365, 366, 368, 369, 370, 371, 372, 373, 376, 377, 378, 379, 380, 381, 382, 383, 385, 386, 387, 388, 389, 390, 392, 394, 395, 398, 399, 400, 401, 402, 404, 405, 407, 408, 409, 410, 412, 413, 414, 415, 416, 417, 418], "outer_test_idx": [10, 12, 15, 31, 35, 42, 46, 52, 53, 54, 62, 66, 67, 71, 73, 75, 77, 81, 89, 90, 92, 98, 99, 105, 123, 124, 141, 144, 146, 154, 155, 164, 169, 176, 178, 182, 189, 190, 195, 198, 201, 206, 216, 217, 218, 219, 231, 234, 244, 247, 248, 258, 262, 270, 277, 278, 279, 286, 288, 291, 292, 301, 304, 306, 316, 319, 325, 328, 343, 357, 359, 362, 367, 374, 375, 384, 391, 393, 396, 397, 403, 406, 411, 419]}

42
models/abl/baseline/20260617_180556/repeat_1/outer_fold_2/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.10285134147020851,
"rmse": 0.3207044456664243,
"mae": 0.2692902485529582,
"r2": -0.10201936510861498
},
"delivery": {
"n_samples": 60,
"mse": 0.663864171054346,
"rmse": 0.8147786024769833,
"mae": 0.548082401487045,
"r2": 0.11628172479876508
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.4457671957671958,
"recall": 0.7063492063492064,
"f1": 0.44368858654572946
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6064246211305034,
"recall": 0.6547619047619048,
"f1": 0.6168530247347982
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3895171114701636,
"js_divergence": 0.10296173296198505
}
}

12
models/abl/baseline/20260617_180556/repeat_1/outer_fold_3/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.1799936812783694,
"lr": 0.00033865324369584493,
"weight_decay": 0.01440133094015265,
"backbone_lr_ratio": 0.3278334190266755,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_1/outer_fold_3/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 4, 5, 7, 8, 9, 10, 11, 12, 14, 15, 16, 17, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 45, 46, 47, 51, 52, 53, 54, 55, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 79, 81, 82, 83, 84, 85, 86, 87, 89, 90, 91, 92, 93, 95, 96, 97, 98, 99, 100, 102, 104, 105, 106, 109, 110, 111, 112, 113, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 127, 129, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 146, 147, 148, 150, 151, 152, 153, 154, 155, 156, 158, 159, 160, 161, 162, 163, 164, 165, 166, 168, 169, 170, 172, 173, 174, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 189, 190, 191, 192, 193, 194, 195, 198, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 212, 213, 214, 215, 216, 217, 218, 219, 221, 223, 224, 226, 227, 228, 230, 231, 233, 234, 235, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 251, 252, 253, 254, 255, 256, 257, 258, 259, 262, 263, 264, 265, 266, 267, 268, 269, 270, 272, 273, 274, 275, 276, 277, 278, 279, 280, 282, 284, 286, 288, 289, 290, 291, 292, 293, 294, 295, 297, 298, 299, 301, 302, 304, 305, 306, 307, 308, 309, 310, 312, 314, 315, 316, 317, 319, 320, 321, 325, 326, 327, 328, 331, 332, 334, 336, 340, 341, 343, 344, 346, 347, 348, 350, 352, 353, 357, 359, 360, 361, 362, 363, 364, 366, 367, 368, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 382, 383, 384, 385, 386, 387, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 400, 401, 402, 403, 404, 406, 407, 410, 411, 412, 414, 415, 416, 418, 419], "outer_test_idx": [3, 6, 13, 18, 19, 26, 33, 44, 48, 49, 50, 56, 57, 72, 80, 88, 94, 101, 103, 107, 108, 114, 126, 128, 130, 145, 149, 157, 167, 171, 175, 188, 196, 197, 199, 211, 220, 222, 225, 229, 232, 236, 250, 260, 261, 271, 281, 283, 285, 287, 296, 300, 303, 311, 313, 318, 322, 323, 324, 329, 330, 333, 335, 337, 338, 339, 342, 345, 349, 351, 354, 355, 356, 358, 365, 369, 381, 388, 399, 405, 408, 409, 413, 417]}

42
models/abl/baseline/20260617_180556/repeat_1/outer_fold_3/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.4783278330356588,
"rmse": 0.6916124876226996,
"mae": 0.3700549229081855,
"r2": -0.4677665928655337
},
"delivery": {
"n_samples": 59,
"mse": 0.8242563169380902,
"rmse": 0.907885629877514,
"mae": 0.6424747443900017,
"r2": 0.19872492929644914
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7023809523809523,
"precision": 0.4810152874709206,
"recall": 0.7291346646185355,
"f1": 0.48990704493096837
},
"ee": {
"n_samples": 84,
"accuracy": 0.6428571428571429,
"precision": 0.5627865961199294,
"recall": 0.5790095525389644,
"f1": 0.5641184865618634
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.40656307520365187,
"js_divergence": 0.10906334576015857
}
}

12
models/abl/baseline/20260617_180556/repeat_1/outer_fold_4/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.14162975892531532,
"lr": 0.0004019907813153378,
"weight_decay": 0.00037960139235180325,
"backbone_lr_ratio": 0.18010931581670617,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_1/outer_fold_4/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 23, 25, 26, 28, 29, 31, 33, 34, 35, 37, 41, 42, 43, 44, 45, 46, 48, 49, 50, 51, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 107, 108, 110, 111, 112, 113, 114, 116, 117, 119, 120, 121, 122, 123, 124, 126, 127, 128, 130, 132, 133, 134, 135, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 149, 152, 154, 155, 157, 160, 161, 162, 163, 164, 165, 166, 167, 169, 170, 171, 173, 174, 175, 176, 177, 178, 180, 181, 182, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 205, 206, 207, 208, 209, 211, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 229, 230, 231, 232, 234, 235, 236, 238, 239, 240, 241, 243, 244, 245, 246, 247, 248, 249, 250, 251, 253, 254, 257, 258, 260, 261, 262, 263, 264, 265, 266, 267, 270, 271, 272, 273, 277, 278, 279, 280, 281, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 296, 298, 299, 300, 301, 302, 303, 304, 305, 306, 309, 311, 312, 313, 314, 315, 316, 318, 319, 320, 322, 323, 324, 325, 326, 328, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 341, 342, 343, 345, 346, 347, 349, 350, 351, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 365, 367, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 381, 383, 384, 385, 386, 388, 390, 391, 392, 393, 394, 395, 396, 397, 399, 400, 401, 402, 403, 405, 406, 407, 408, 409, 410, 411, 413, 415, 416, 417, 418, 419], "outer_test_idx": [0, 5, 11, 21, 24, 27, 30, 32, 36, 38, 39, 40, 47, 55, 58, 61, 65, 76, 86, 95, 100, 106, 109, 115, 118, 125, 129, 131, 136, 148, 150, 151, 153, 156, 158, 159, 168, 172, 179, 183, 184, 185, 202, 204, 210, 212, 213, 228, 233, 237, 242, 252, 255, 256, 259, 268, 269, 274, 275, 276, 282, 297, 307, 308, 310, 317, 321, 327, 340, 344, 348, 352, 364, 366, 368, 379, 380, 382, 387, 389, 398, 404, 412, 414]}

42
models/abl/baseline/20260617_180556/repeat_1/outer_fold_4/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.16667444743348794,
"rmse": 0.4082578198069058,
"mae": 0.2857146007674081,
"r2": -0.20998938958412183
},
"delivery": {
"n_samples": 58,
"mse": 0.9768153651071599,
"rmse": 0.9883397012703476,
"mae": 0.6390594764259355,
"r2": 0.055435667309009284
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.49700048379293665,
"recall": 0.7552483358934973,
"f1": 0.533806146572104
},
"ee": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.6299589603283173,
"recall": 0.6871794871794873,
"f1": 0.6483638563493374
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.394783094910233,
"js_divergence": 0.10592993993751994
}
}

12
models/abl/baseline/20260617_180556/repeat_2/outer_fold_0/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.28242799368681437,
"lr": 0.00037183641805732076,
"weight_decay": 6.290644294586152e-05,
"backbone_lr_ratio": 0.10677482709481352,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_2/outer_fold_0/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72, 73, 75, 76, 77, 80, 81, 85, 86, 88, 89, 90, 92, 94, 95, 96, 98, 99, 100, 101, 103, 104, 105, 106, 107, 108, 109, 111, 113, 114, 115, 116, 117, 118, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 133, 134, 135, 136, 138, 139, 141, 144, 145, 146, 147, 148, 149, 150, 151, 153, 154, 155, 156, 157, 158, 159, 160, 163, 164, 167, 168, 169, 170, 171, 172, 175, 176, 178, 179, 180, 182, 183, 184, 185, 188, 189, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 206, 208, 210, 211, 212, 213, 216, 217, 218, 219, 220, 222, 224, 225, 227, 228, 229, 230, 231, 232, 233, 234, 236, 237, 239, 241, 242, 244, 245, 247, 248, 249, 250, 252, 253, 255, 256, 257, 258, 259, 260, 261, 262, 265, 267, 268, 269, 270, 271, 272, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 285, 286, 287, 288, 291, 292, 294, 296, 297, 298, 299, 300, 301, 303, 304, 306, 307, 308, 309, 310, 311, 312, 313, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 333, 335, 336, 337, 338, 339, 340, 342, 343, 344, 345, 346, 347, 348, 349, 351, 352, 354, 355, 356, 357, 358, 359, 360, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372, 374, 375, 377, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 393, 395, 396, 397, 398, 399, 400, 401, 403, 404, 405, 406, 407, 408, 409, 411, 412, 413, 414, 415, 417, 418, 419], "outer_test_idx": [7, 16, 20, 29, 51, 59, 60, 69, 70, 74, 78, 79, 82, 83, 84, 87, 91, 93, 97, 102, 110, 112, 119, 120, 121, 132, 137, 140, 142, 143, 152, 161, 162, 165, 166, 173, 174, 177, 181, 186, 187, 194, 205, 207, 209, 214, 215, 221, 223, 226, 235, 238, 240, 243, 246, 251, 254, 263, 264, 266, 273, 284, 289, 290, 293, 295, 302, 305, 314, 315, 331, 334, 341, 350, 353, 361, 373, 376, 378, 392, 394, 402, 410, 416]}

42
models/abl/baseline/20260617_180556/repeat_2/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.38432886654317966,
"rmse": 0.6199426316548812,
"mae": 0.43342806632260245,
"r2": -0.2761097384806048
},
"delivery": {
"n_samples": 58,
"mse": 0.6744019580730524,
"rmse": 0.8212197988803317,
"mae": 0.6701494166306381,
"r2": 0.14279441523317304
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.35634973404255316,
"recall": 0.547423887587822,
"f1": 0.3495414294750159
},
"ee": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.6587301587301587,
"recall": 0.6938095238095238,
"f1": 0.6526081185986528
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.40697274862376626,
"js_divergence": 0.09501893155280818
}
}

12
models/abl/baseline/20260617_180556/repeat_2/outer_fold_1/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.23085562232445592,
"lr": 0.0005227270589225132,
"weight_decay": 0.00461701039547356,
"backbone_lr_ratio": 0.12087164274488635,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_2/outer_fold_1/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 3, 5, 6, 7, 10, 11, 12, 13, 15, 16, 18, 19, 20, 21, 24, 26, 27, 29, 30, 31, 32, 33, 35, 36, 38, 39, 40, 42, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 97, 98, 99, 100, 101, 102, 103, 105, 106, 107, 108, 109, 110, 112, 114, 115, 118, 119, 120, 121, 123, 124, 125, 126, 128, 129, 130, 131, 132, 136, 137, 140, 141, 142, 143, 144, 145, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 161, 162, 164, 165, 166, 167, 168, 169, 171, 172, 173, 174, 175, 176, 177, 178, 179, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 194, 195, 196, 197, 198, 199, 201, 202, 204, 205, 206, 207, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 244, 246, 247, 248, 250, 251, 252, 254, 255, 256, 258, 259, 260, 261, 262, 263, 264, 266, 268, 269, 270, 271, 273, 274, 275, 276, 277, 278, 279, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 300, 301, 302, 303, 304, 305, 306, 307, 308, 310, 311, 313, 314, 315, 316, 317, 318, 319, 321, 322, 323, 324, 325, 327, 328, 329, 330, 331, 333, 334, 335, 337, 338, 339, 340, 341, 342, 343, 344, 345, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 361, 362, 364, 365, 366, 367, 368, 369, 373, 374, 375, 376, 378, 379, 380, 381, 382, 384, 387, 388, 389, 391, 392, 393, 394, 396, 397, 398, 399, 402, 403, 404, 405, 406, 408, 409, 410, 411, 412, 413, 414, 416, 417, 419], "outer_test_idx": [1, 2, 4, 8, 9, 14, 17, 22, 23, 25, 28, 34, 37, 41, 43, 45, 63, 64, 68, 85, 96, 104, 111, 113, 116, 117, 122, 127, 133, 134, 135, 138, 139, 147, 160, 163, 170, 180, 191, 192, 193, 200, 203, 208, 224, 227, 230, 239, 241, 245, 249, 253, 257, 265, 267, 272, 280, 294, 298, 299, 309, 312, 320, 326, 332, 336, 346, 347, 360, 363, 370, 371, 372, 377, 383, 385, 386, 390, 395, 400, 401, 407, 415, 418]}

42
models/abl/baseline/20260617_180556/repeat_2/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.28709899653313253,
"rmse": 0.5358161965946275,
"mae": 0.41961162147067843,
"r2": -3.7562675545651993
},
"delivery": {
"n_samples": 61,
"mse": 1.0055884332484593,
"rmse": 1.0027903236711349,
"mae": 0.6592718309371687,
"r2": 0.23884204782685936
},
"pdi": {
"n_samples": 84,
"accuracy": 0.44047619047619047,
"precision": 0.34375,
"recall": 0.32275132275132273,
"f1": 0.2913806254767353
},
"ee": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.5,
"recall": 0.5004995004995004,
"f1": 0.4796109240553685
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.4974494276121329,
"js_divergence": 0.1259321573319038
}
}

12
models/abl/baseline/20260617_180556/repeat_2/outer_fold_2/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.18117503720006686,
"lr": 8.135394049241399e-05,
"weight_decay": 0.015076469897649124,
"backbone_lr_ratio": 0.9586175198423679,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_2/outer_fold_2/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 36, 37, 38, 39, 40, 41, 43, 44, 45, 47, 48, 49, 50, 51, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 68, 69, 70, 72, 74, 76, 78, 79, 80, 82, 83, 84, 85, 86, 87, 88, 91, 93, 94, 95, 96, 97, 100, 101, 102, 103, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 142, 143, 145, 147, 148, 149, 150, 151, 152, 153, 156, 157, 158, 159, 160, 161, 162, 163, 165, 166, 167, 168, 170, 171, 172, 173, 174, 175, 177, 179, 180, 181, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 196, 197, 199, 200, 202, 203, 204, 205, 207, 208, 209, 210, 211, 212, 213, 214, 215, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 232, 233, 235, 236, 237, 238, 239, 240, 241, 242, 243, 245, 246, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 264, 265, 266, 267, 268, 269, 271, 272, 273, 274, 275, 276, 280, 281, 282, 283, 284, 285, 287, 289, 290, 293, 294, 295, 296, 297, 298, 299, 300, 302, 303, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 317, 318, 320, 321, 322, 323, 324, 326, 327, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 358, 360, 361, 363, 364, 365, 366, 368, 369, 370, 371, 372, 373, 376, 377, 378, 379, 380, 381, 382, 383, 385, 386, 387, 388, 389, 390, 392, 394, 395, 398, 399, 400, 401, 402, 404, 405, 407, 408, 409, 410, 412, 413, 414, 415, 416, 417, 418], "outer_test_idx": [10, 12, 15, 31, 35, 42, 46, 52, 53, 54, 62, 66, 67, 71, 73, 75, 77, 81, 89, 90, 92, 98, 99, 105, 123, 124, 141, 144, 146, 154, 155, 164, 169, 176, 178, 182, 189, 190, 195, 198, 201, 206, 216, 217, 218, 219, 231, 234, 244, 247, 248, 258, 262, 270, 277, 278, 279, 286, 288, 291, 292, 301, 304, 306, 316, 319, 325, 328, 343, 357, 359, 362, 367, 374, 375, 384, 391, 393, 396, 397, 403, 406, 411, 419]}

42
models/abl/baseline/20260617_180556/repeat_2/outer_fold_2/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.12734599676966307,
"rmse": 0.3568557086129674,
"mae": 0.283481532619113,
"r2": -0.3644717949534717
},
"delivery": {
"n_samples": 60,
"mse": 0.6995736506531275,
"rmse": 0.8364051952571359,
"mae": 0.5743296845893686,
"r2": 0.06874621814647808
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5119047619047619,
"precision": 0.41008771929824556,
"recall": 0.6746031746031745,
"f1": 0.4050345260514752
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6189974937343359,
"recall": 0.687908496732026,
"f1": 0.6319077437787746
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3964743242871827,
"js_divergence": 0.10757710661599443
}
}

12
models/abl/baseline/20260617_180556/repeat_2/outer_fold_3/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.1799936812783694,
"lr": 0.00033865324369584493,
"weight_decay": 0.01440133094015265,
"backbone_lr_ratio": 0.3278334190266755,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_2/outer_fold_3/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 4, 5, 7, 8, 9, 10, 11, 12, 14, 15, 16, 17, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 45, 46, 47, 51, 52, 53, 54, 55, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 79, 81, 82, 83, 84, 85, 86, 87, 89, 90, 91, 92, 93, 95, 96, 97, 98, 99, 100, 102, 104, 105, 106, 109, 110, 111, 112, 113, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 127, 129, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 146, 147, 148, 150, 151, 152, 153, 154, 155, 156, 158, 159, 160, 161, 162, 163, 164, 165, 166, 168, 169, 170, 172, 173, 174, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 189, 190, 191, 192, 193, 194, 195, 198, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 212, 213, 214, 215, 216, 217, 218, 219, 221, 223, 224, 226, 227, 228, 230, 231, 233, 234, 235, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 251, 252, 253, 254, 255, 256, 257, 258, 259, 262, 263, 264, 265, 266, 267, 268, 269, 270, 272, 273, 274, 275, 276, 277, 278, 279, 280, 282, 284, 286, 288, 289, 290, 291, 292, 293, 294, 295, 297, 298, 299, 301, 302, 304, 305, 306, 307, 308, 309, 310, 312, 314, 315, 316, 317, 319, 320, 321, 325, 326, 327, 328, 331, 332, 334, 336, 340, 341, 343, 344, 346, 347, 348, 350, 352, 353, 357, 359, 360, 361, 362, 363, 364, 366, 367, 368, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 382, 383, 384, 385, 386, 387, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 400, 401, 402, 403, 404, 406, 407, 410, 411, 412, 414, 415, 416, 418, 419], "outer_test_idx": [3, 6, 13, 18, 19, 26, 33, 44, 48, 49, 50, 56, 57, 72, 80, 88, 94, 101, 103, 107, 108, 114, 126, 128, 130, 145, 149, 157, 167, 171, 175, 188, 196, 197, 199, 211, 220, 222, 225, 229, 232, 236, 250, 260, 261, 271, 281, 283, 285, 287, 296, 300, 303, 311, 313, 318, 322, 323, 324, 329, 330, 333, 335, 337, 338, 339, 342, 345, 349, 351, 354, 355, 356, 358, 365, 369, 381, 388, 399, 405, 408, 409, 413, 417]}

42
models/abl/baseline/20260617_180556/repeat_2/outer_fold_3/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.46358638894813786,
"rmse": 0.6808717859833361,
"mae": 0.37019611553973464,
"r2": -0.422531928127458
},
"delivery": {
"n_samples": 59,
"mse": 0.8370140113507513,
"rmse": 0.9148846983914155,
"mae": 0.6370043654657774,
"r2": 0.18632293457411198
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6071428571428571,
"precision": 0.47548510048510045,
"recall": 0.7491039426523297,
"f1": 0.4726173766801967
},
"ee": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.6115219070742745,
"recall": 0.6469582704876823,
"f1": 0.6198067632850242
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.40801997248379873,
"js_divergence": 0.11086638268723202
}
}

12
models/abl/baseline/20260617_180556/repeat_2/outer_fold_4/best_params.json
View File

@ -1,12 +0,0 @@
{
"dropout": 0.14162975892531532,
"lr": 0.0004019907813153378,
"weight_decay": 0.00037960139235180325,
"backbone_lr_ratio": 0.18010931581670617,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/baseline/20260617_180556/repeat_2/outer_fold_4/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 23, 25, 26, 28, 29, 31, 33, 34, 35, 37, 41, 42, 43, 44, 45, 46, 48, 49, 50, 51, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 107, 108, 110, 111, 112, 113, 114, 116, 117, 119, 120, 121, 122, 123, 124, 126, 127, 128, 130, 132, 133, 134, 135, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 149, 152, 154, 155, 157, 160, 161, 162, 163, 164, 165, 166, 167, 169, 170, 171, 173, 174, 175, 176, 177, 178, 180, 181, 182, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 205, 206, 207, 208, 209, 211, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 229, 230, 231, 232, 234, 235, 236, 238, 239, 240, 241, 243, 244, 245, 246, 247, 248, 249, 250, 251, 253, 254, 257, 258, 260, 261, 262, 263, 264, 265, 266, 267, 270, 271, 272, 273, 277, 278, 279, 280, 281, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 296, 298, 299, 300, 301, 302, 303, 304, 305, 306, 309, 311, 312, 313, 314, 315, 316, 318, 319, 320, 322, 323, 324, 325, 326, 328, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 341, 342, 343, 345, 346, 347, 349, 350, 351, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 365, 367, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 381, 383, 384, 385, 386, 388, 390, 391, 392, 393, 394, 395, 396, 397, 399, 400, 401, 402, 403, 405, 406, 407, 408, 409, 410, 411, 413, 415, 416, 417, 418, 419], "outer_test_idx": [0, 5, 11, 21, 24, 27, 30, 32, 36, 38, 39, 40, 47, 55, 58, 61, 65, 76, 86, 95, 100, 106, 109, 115, 118, 125, 129, 131, 136, 148, 150, 151, 153, 156, 158, 159, 168, 172, 179, 183, 184, 185, 202, 204, 210, 212, 213, 228, 233, 237, 242, 252, 255, 256, 259, 268, 269, 274, 275, 276, 282, 297, 307, 308, 310, 317, 321, 327, 340, 344, 348, 352, 364, 366, 368, 379, 380, 382, 387, 389, 398, 404, 412, 414]}

42
models/abl/baseline/20260617_180556/repeat_2/outer_fold_4/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.15145313313910996,
"rmse": 0.3891697998806048,
"mae": 0.2744193644750686,
"r2": -0.09948877551085711
},
"delivery": {
"n_samples": 58,
"mse": 0.9378577638427679,
"rmse": 0.9684305673835207,
"mae": 0.6135335185661398,
"r2": 0.09310702461562304
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6071428571428571,
"precision": 0.4863247863247863,
"recall": 0.7491039426523297,
"f1": 0.5019607843137255
},
"ee": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.6324786324786325,
"recall": 0.7021367521367522,
"f1": 0.6488452445899254
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.39116454524118993,
"js_divergence": 0.1065009133663494
}
}

932
models/abl/baseline/20260617_180556/summary.json
View File

@ -1,932 +0,0 @@
{
"fold_results": [
{
"fold": 0,
"best_params": {
"dropout": 0.28242799368681437,
"lr": 0.00037183641805732076,
"weight_decay": 6.290644294586152e-05,
"backbone_lr_ratio": 0.10677482709481352,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 15,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.32790454280151365,
"rmse": 0.572629498717551,
"mae": 0.35871042688208893,
"r2": -0.08876073797082196
},
"delivery": {
"n_samples": 58,
"mse": 0.7250728333301815,
"rmse": 0.8515120864263651,
"mae": 0.7098816664696768,
"r2": 0.07838867510225656
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.3435971685971686,
"recall": 0.5359094457455114,
"f1": 0.34977324263038545
},
"ee": {
"n_samples": 84,
"accuracy": 0.7142857142857143,
"precision": 0.6871101871101871,
"recall": 0.707142857142857,
"f1": 0.6756066411238826
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.4073394583325288,
"js_divergence": 0.092887269703335
}
}
},
{
"fold": 1,
"best_params": {
"dropout": 0.23085562232445592,
"lr": 0.0005227270589225132,
"weight_decay": 0.00461701039547356,
"backbone_lr_ratio": 0.12087164274488635,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 9,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.233876145855699,
"rmse": 0.4836074294876982,
"mae": 0.3760785318556286,
"r2": -2.874543407510119
},
"delivery": {
"n_samples": 61,
"mse": 1.0238964615450643,
"rmse": 1.0118776910007772,
"mae": 0.6578027546466862,
"r2": 0.2249841902125319
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5238095238095238,
"precision": 0.34497354497354493,
"recall": 0.3280423280423281,
"f1": 0.3231922398589065
},
"ee": {
"n_samples": 84,
"accuracy": 0.5595238095238095,
"precision": 0.5101731601731602,
"recall": 0.5036154321868608,
"f1": 0.48790324563862697
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.48528378254405913,
"js_divergence": 0.12258377401744426
}
}
},
{
"fold": 2,
"best_params": {
"dropout": 0.18117503720006686,
"lr": 8.135394049241399e-05,
"weight_decay": 0.015076469897649124,
"backbone_lr_ratio": 0.9586175198423679,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 22,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.07870736214858826,
"rmse": 0.28054832408800495,
"mae": 0.2252055633635748,
"r2": 0.15667568332606574
},
"delivery": {
"n_samples": 60,
"mse": 0.6624288051399898,
"rmse": 0.8138972939750997,
"mae": 0.5578871982870623,
"r2": 0.118192445613993
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5714285714285714,
"precision": 0.43453195231581004,
"recall": 0.7014848950332823,
"f1": 0.4452380952380952
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6140109890109889,
"recall": 0.6785714285714285,
"f1": 0.6262626262626263
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3967825471495568,
"js_divergence": 0.1057568924841401
}
}
},
{
"fold": 3,
"best_params": {
"dropout": 0.1799936812783694,
"lr": 0.00033865324369584493,
"weight_decay": 0.01440133094015265,
"backbone_lr_ratio": 0.3278334190266755,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 10,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.477705869562307,
"rmse": 0.6911626939891266,
"mae": 0.38659381004701177,
"r2": -0.46585807501413723
},
"delivery": {
"n_samples": 59,
"mse": 0.7692103143544243,
"rmse": 0.8770463581558413,
"mae": 0.6155587352528158,
"r2": 0.2522361838732067
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.4426767676767677,
"recall": 0.6817716333845366,
"f1": 0.43613724611708476
},
"ee": {
"n_samples": 84,
"accuracy": 0.6428571428571429,
"precision": 0.5650987224157956,
"recall": 0.5790095525389644,
"f1": 0.5637508626639062
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.4102080974965646,
"js_divergence": 0.11100613154873733
}
}
},
{
"fold": 4,
"best_params": {
"dropout": 0.14162975892531532,
"lr": 0.0004019907813153378,
"weight_decay": 0.00037960139235180325,
"backbone_lr_ratio": 0.18010931581670617,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 12,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.1842198466764517,
"rmse": 0.4292083953937198,
"mae": 0.29178874833243235,
"r2": -0.33736192476816806
},
"delivery": {
"n_samples": 58,
"mse": 0.9003281069383875,
"rmse": 0.9488562098328637,
"mae": 0.6070586243837044,
"r2": 0.1293975832987566
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.34548450889914306,
"recall": 0.5337941628264209,
"f1": 0.3497882046269143
},
"ee": {
"n_samples": 84,
"accuracy": 0.7023809523809523,
"precision": 0.6548765244417418,
"recall": 0.7576923076923077,
"f1": 0.6777508856852221
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3946735882065513,
"js_divergence": 0.1067245125767987
}
}
},
{
"fold": 0,
"best_params": {
"dropout": 0.28242799368681437,
"lr": 0.00037183641805732076,
"weight_decay": 6.290644294586152e-05,
"backbone_lr_ratio": 0.10677482709481352,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 15,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.3808128541133737,
"rmse": 0.6171003598389598,
"mae": 0.4168719153806388,
"r2": -0.2644353156285024
},
"delivery": {
"n_samples": 58,
"mse": 0.751610048614176,
"rmse": 0.8669544674400012,
"mae": 0.7122741444999802,
"r2": 0.04465827311677528
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5,
"precision": 0.3181788931788932,
"recall": 0.5039032006245121,
"f1": 0.32369614512471656
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6258687258687259,
"recall": 0.6461904761904762,
"f1": 0.6127803231251508
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.39644445917049376,
"js_divergence": 0.09334798400608228
}
}
},
{
"fold": 1,
"best_params": {
"dropout": 0.23085562232445592,
"lr": 0.0005227270589225132,
"weight_decay": 0.00461701039547356,
"backbone_lr_ratio": 0.12087164274488635,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 9,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.3074589292084139,
"rmse": 0.5544897917982025,
"mae": 0.3793355481965201,
"r2": -4.093563359726248
},
"delivery": {
"n_samples": 61,
"mse": 1.0592195690639976,
"rmse": 1.029183933543464,
"mae": 0.6786946221025753,
"r2": 0.19824714422578682
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5952380952380952,
"precision": 0.37142213207121927,
"recall": 0.35978835978835977,
"f1": 0.35976608187134507
},
"ee": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.5052287581699346,
"recall": 0.5094429380143666,
"f1": 0.4903811048389362
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.5219711265981701,
"js_divergence": 0.13035793372268764
}
}
},
{
"fold": 2,
"best_params": {
"dropout": 0.18117503720006686,
"lr": 8.135394049241399e-05,
"weight_decay": 0.015076469897649124,
"backbone_lr_ratio": 0.9586175198423679,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 22,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.10285134147020851,
"rmse": 0.3207044456664243,
"mae": 0.2692902485529582,
"r2": -0.10201936510861498
},
"delivery": {
"n_samples": 60,
"mse": 0.663864171054346,
"rmse": 0.8147786024769833,
"mae": 0.548082401487045,
"r2": 0.11628172479876508
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.4457671957671958,
"recall": 0.7063492063492064,
"f1": 0.44368858654572946
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6064246211305034,
"recall": 0.6547619047619048,
"f1": 0.6168530247347982
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3895171114701636,
"js_divergence": 0.10296173296198505
}
}
},
{
"fold": 3,
"best_params": {
"dropout": 0.1799936812783694,
"lr": 0.00033865324369584493,
"weight_decay": 0.01440133094015265,
"backbone_lr_ratio": 0.3278334190266755,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 10,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.4783278330356588,
"rmse": 0.6916124876226996,
"mae": 0.3700549229081855,
"r2": -0.4677665928655337
},
"delivery": {
"n_samples": 59,
"mse": 0.8242563169380902,
"rmse": 0.907885629877514,
"mae": 0.6424747443900017,
"r2": 0.19872492929644914
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7023809523809523,
"precision": 0.4810152874709206,
"recall": 0.7291346646185355,
"f1": 0.48990704493096837
},
"ee": {
"n_samples": 84,
"accuracy": 0.6428571428571429,
"precision": 0.5627865961199294,
"recall": 0.5790095525389644,
"f1": 0.5641184865618634
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.40656307520365187,
"js_divergence": 0.10906334576015857
}
}
},
{
"fold": 4,
"best_params": {
"dropout": 0.14162975892531532,
"lr": 0.0004019907813153378,
"weight_decay": 0.00037960139235180325,
"backbone_lr_ratio": 0.18010931581670617,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 12,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.16667444743348794,
"rmse": 0.4082578198069058,
"mae": 0.2857146007674081,
"r2": -0.20998938958412183
},
"delivery": {
"n_samples": 58,
"mse": 0.9768153651071599,
"rmse": 0.9883397012703476,
"mae": 0.6390594764259355,
"r2": 0.055435667309009284
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.49700048379293665,
"recall": 0.7552483358934973,
"f1": 0.533806146572104
},
"ee": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.6299589603283173,
"recall": 0.6871794871794873,
"f1": 0.6483638563493374
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.394783094910233,
"js_divergence": 0.10592993993751994
}
}
},
{
"fold": 0,
"best_params": {
"dropout": 0.28242799368681437,
"lr": 0.00037183641805732076,
"weight_decay": 6.290644294586152e-05,
"backbone_lr_ratio": 0.10677482709481352,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 15,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.38432886654317966,
"rmse": 0.6199426316548812,
"mae": 0.43342806632260245,
"r2": -0.2761097384806048
},
"delivery": {
"n_samples": 58,
"mse": 0.6744019580730524,
"rmse": 0.8212197988803317,
"mae": 0.6701494166306381,
"r2": 0.14279441523317304
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.35634973404255316,
"recall": 0.547423887587822,
"f1": 0.3495414294750159
},
"ee": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.6587301587301587,
"recall": 0.6938095238095238,
"f1": 0.6526081185986528
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.40697274862376626,
"js_divergence": 0.09501893155280818
}
}
},
{
"fold": 1,
"best_params": {
"dropout": 0.23085562232445592,
"lr": 0.0005227270589225132,
"weight_decay": 0.00461701039547356,
"backbone_lr_ratio": 0.12087164274488635,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 9,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.28709899653313253,
"rmse": 0.5358161965946275,
"mae": 0.41961162147067843,
"r2": -3.7562675545651993
},
"delivery": {
"n_samples": 61,
"mse": 1.0055884332484593,
"rmse": 1.0027903236711349,
"mae": 0.6592718309371687,
"r2": 0.23884204782685936
},
"pdi": {
"n_samples": 84,
"accuracy": 0.44047619047619047,
"precision": 0.34375,
"recall": 0.32275132275132273,
"f1": 0.2913806254767353
},
"ee": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.5,
"recall": 0.5004995004995004,
"f1": 0.4796109240553685
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.4974494276121329,
"js_divergence": 0.1259321573319038
}
}
},
{
"fold": 2,
"best_params": {
"dropout": 0.18117503720006686,
"lr": 8.135394049241399e-05,
"weight_decay": 0.015076469897649124,
"backbone_lr_ratio": 0.9586175198423679,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 22,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.12734599676966307,
"rmse": 0.3568557086129674,
"mae": 0.283481532619113,
"r2": -0.3644717949534717
},
"delivery": {
"n_samples": 60,
"mse": 0.6995736506531275,
"rmse": 0.8364051952571359,
"mae": 0.5743296845893686,
"r2": 0.06874621814647808
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5119047619047619,
"precision": 0.41008771929824556,
"recall": 0.6746031746031745,
"f1": 0.4050345260514752
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6189974937343359,
"recall": 0.687908496732026,
"f1": 0.6319077437787746
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3964743242871827,
"js_divergence": 0.10757710661599443
}
}
},
{
"fold": 3,
"best_params": {
"dropout": 0.1799936812783694,
"lr": 0.00033865324369584493,
"weight_decay": 0.01440133094015265,
"backbone_lr_ratio": 0.3278334190266755,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 10,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.46358638894813786,
"rmse": 0.6808717859833361,
"mae": 0.37019611553973464,
"r2": -0.422531928127458
},
"delivery": {
"n_samples": 59,
"mse": 0.8370140113507513,
"rmse": 0.9148846983914155,
"mae": 0.6370043654657774,
"r2": 0.18632293457411198
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6071428571428571,
"precision": 0.47548510048510045,
"recall": 0.7491039426523297,
"f1": 0.4726173766801967
},
"ee": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.6115219070742745,
"recall": 0.6469582704876823,
"f1": 0.6198067632850242
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.40801997248379873,
"js_divergence": 0.11086638268723202
}
}
},
{
"fold": 4,
"best_params": {
"dropout": 0.14162975892531532,
"lr": 0.0004019907813153378,
"weight_decay": 0.00037960139235180325,
"backbone_lr_ratio": 0.18010931581670617,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 12,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.15145313313910996,
"rmse": 0.3891697998806048,
"mae": 0.2744193644750686,
"r2": -0.09948877551085711
},
"delivery": {
"n_samples": 58,
"mse": 0.9378577638427679,
"rmse": 0.9684305673835207,
"mae": 0.6135335185661398,
"r2": 0.09310702461562304
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6071428571428571,
"precision": 0.4863247863247863,
"recall": 0.7491039426523297,
"f1": 0.5019607843137255
},
"ee": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.6324786324786325,
"recall": 0.7021367521367522,
"f1": 0.6488452445899254
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.39116454524118993,
"js_divergence": 0.1065009133663494
}
}
}
],
"summary_stats": {
"size": {
"mse_mean": 0.2768235036159284,
"mse_std": 0.1346583260723207,
"rmse_mean": 0.508798491275714,
"rmse_std": 0.13396864891266763,
"mae_mean": 0.3427187344475763,
"mae_std": 0.06271329639776796,
"r2_mean": -0.9110994850991861,
"r2_std": 1.3609378589681
},
"delivery": {
"mse_mean": 0.834075853950265,
"mse_std": 0.13581578656540713,
"rmse_mean": 0.9102708371721865,
"rmse_std": 0.0740463162089242,
"mae_mean": 0.634870878942305,
"mae_std": 0.04810398332825185,
"r2_mean": 0.14309063048291837,
"r2_std": 0.06678544282427759
},
"pdi": {
"accuracy_mean": 0.573809523809524,
"accuracy_std": 0.07315376902732006,
"precision_mean": 0.40644301832628565,
"precision_std": 0.06081224559746048,
"recall_mean": 0.591894166836878,
"recall_std": 0.15212645465643837,
"f1_mean": 0.40503518503422653,
"f1_std": 0.07270554354315185
},
"ee": {
"accuracy_mean": 0.65,
"accuracy_std": 0.05460640448180816,
"precision_mean": 0.5988843624524457,
"precision_std": 0.055997857124168736,
"recall_mean": 0.6355952320322067,
"recall_std": 0.07902933872282923,
"f1_mean": 0.5997699900861396,
"f1_std": 0.06505269720858664
},
"toxic": {
"accuracy_mean": 0.9500396342534485,
"accuracy_std": 0.014290090778642328,
"precision_mean": 0.7457142857142857,
"precision_std": 0.03149343955006944,
"recall_mean": 0.9737706334802525,
"recall_std": 0.007575117962473199,
"f1_mean": 0.81483431023254,
"f1_std": 0.0312513648447909
},
"biodist": {
"kl_divergence_mean": 0.42024315728866957,
"kl_divergence_std": 0.04170766696370389,
"js_divergence_mean": 0.10843433388487847,
"js_divergence_std": 0.010645869413863087
}
}
}

16
models/abl/both/20260617_232841/outer_fold_0/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.12930599750984867,
"lr": 0.0003183893957684731,
"weight_decay": 0.0011797820449629415,
"backbone_lr_ratio": 0.35598913444921026,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/outer_fold_0/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72, 73, 75, 76, 77, 80, 81, 85, 86, 88, 89, 90, 92, 94, 95, 96, 98, 99, 100, 101, 103, 104, 105, 106, 107, 108, 109, 111, 113, 114, 115, 116, 117, 118, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 133, 134, 135, 136, 138, 139, 141, 144, 145, 146, 147, 148, 149, 150, 151, 153, 154, 155, 156, 157, 158, 159, 160, 163, 164, 167, 168, 169, 170, 171, 172, 175, 176, 178, 179, 180, 182, 183, 184, 185, 188, 189, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 206, 208, 210, 211, 212, 213, 216, 217, 218, 219, 220, 222, 224, 225, 227, 228, 229, 230, 231, 232, 233, 234, 236, 237, 239, 241, 242, 244, 245, 247, 248, 249, 250, 252, 253, 255, 256, 257, 258, 259, 260, 261, 262, 265, 267, 268, 269, 270, 271, 272, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 285, 286, 287, 288, 291, 292, 294, 296, 297, 298, 299, 300, 301, 303, 304, 306, 307, 308, 309, 310, 311, 312, 313, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 333, 335, 336, 337, 338, 339, 340, 342, 343, 344, 345, 346, 347, 348, 349, 351, 352, 354, 355, 356, 357, 358, 359, 360, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372, 374, 375, 377, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 393, 395, 396, 397, 398, 399, 400, 401, 403, 404, 405, 406, 407, 408, 409, 411, 412, 413, 414, 415, 417, 418, 419], "outer_test_idx": [7, 16, 20, 29, 51, 59, 60, 69, 70, 74, 78, 79, 82, 83, 84, 87, 91, 93, 97, 102, 110, 112, 119, 120, 121, 132, 137, 140, 142, 143, 152, 161, 162, 165, 166, 173, 174, 177, 181, 186, 187, 194, 205, 207, 209, 214, 215, 221, 223, 226, 235, 238, 240, 243, 246, 251, 254, 263, 264, 266, 273, 284, 289, 290, 293, 295, 302, 305, 314, 315, 331, 334, 341, 350, 353, 361, 373, 376, 378, 392, 394, 402, 410, 416]}

42
models/abl/both/20260617_232841/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.27316252507783767,
"rmse": 0.5226495241343262,
"mae": 0.268634506018765,
"r2": 0.09300240292875017
},
"delivery": {
"n_samples": 58,
"mse": 0.49318943105662066,
"rmse": 0.7022744698881063,
"mae": 0.5225495728182381,
"r2": 0.37312647214479966
},
"pdi": {
"n_samples": 84,
"accuracy": 0.75,
"precision": 0.4107142857142857,
"recall": 0.5977751756440282,
"f1": 0.447008547008547
},
"ee": {
"n_samples": 84,
"accuracy": 0.7023809523809523,
"precision": 0.6658119658119658,
"recall": 0.6976190476190475,
"f1": 0.6690688363683327
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.20575334902695963,
"js_divergence": 0.045069148869594025
}
}

16
models/abl/both/20260617_232841/outer_fold_1/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.14959924395216218,
"lr": 0.00012401964324838103,
"weight_decay": 9.656428689903883e-05,
"backbone_lr_ratio": 0.32683834688760083,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/outer_fold_1/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 3, 5, 6, 7, 10, 11, 12, 13, 15, 16, 18, 19, 20, 21, 24, 26, 27, 29, 30, 31, 32, 33, 35, 36, 38, 39, 40, 42, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 97, 98, 99, 100, 101, 102, 103, 105, 106, 107, 108, 109, 110, 112, 114, 115, 118, 119, 120, 121, 123, 124, 125, 126, 128, 129, 130, 131, 132, 136, 137, 140, 141, 142, 143, 144, 145, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 161, 162, 164, 165, 166, 167, 168, 169, 171, 172, 173, 174, 175, 176, 177, 178, 179, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 194, 195, 196, 197, 198, 199, 201, 202, 204, 205, 206, 207, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 244, 246, 247, 248, 250, 251, 252, 254, 255, 256, 258, 259, 260, 261, 262, 263, 264, 266, 268, 269, 270, 271, 273, 274, 275, 276, 277, 278, 279, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 300, 301, 302, 303, 304, 305, 306, 307, 308, 310, 311, 313, 314, 315, 316, 317, 318, 319, 321, 322, 323, 324, 325, 327, 328, 329, 330, 331, 333, 334, 335, 337, 338, 339, 340, 341, 342, 343, 344, 345, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 361, 362, 364, 365, 366, 367, 368, 369, 373, 374, 375, 376, 378, 379, 380, 381, 382, 384, 387, 388, 389, 391, 392, 393, 394, 396, 397, 398, 399, 402, 403, 404, 405, 406, 408, 409, 410, 411, 412, 413, 414, 416, 417, 419], "outer_test_idx": [1, 2, 4, 8, 9, 14, 17, 22, 23, 25, 28, 34, 37, 41, 43, 45, 63, 64, 68, 85, 96, 104, 111, 113, 116, 117, 122, 127, 133, 134, 135, 138, 139, 147, 160, 163, 170, 180, 191, 192, 193, 200, 203, 208, 224, 227, 230, 239, 241, 245, 249, 253, 257, 265, 267, 272, 280, 294, 298, 299, 309, 312, 320, 326, 332, 336, 346, 347, 360, 363, 370, 371, 372, 377, 383, 385, 386, 390, 395, 400, 401, 407, 415, 418]}

42
models/abl/both/20260617_232841/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.11199287760383045,
"rmse": 0.3346533693298641,
"mae": 0.24370252518426805,
"r2": -0.8553464014912291
},
"delivery": {
"n_samples": 61,
"mse": 0.9945337433308287,
"rmse": 0.9972631264269368,
"mae": 0.637184505488296,
"r2": 0.24720965117371863
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5833333333333334,
"precision": 0.2606719367588933,
"recall": 0.24206349206349206,
"f1": 0.25038520801232667
},
"ee": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.4737424924924925,
"recall": 0.47331240188383045,
"f1": 0.46359870390878144
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.42904510824573017,
"js_divergence": 0.11140172525390164
}
}

16
models/abl/both/20260617_232841/outer_fold_2/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.12302210383900847,
"lr": 0.00015510064533248203,
"weight_decay": 0.0037340196924722196,
"backbone_lr_ratio": 0.3036603338505521,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/outer_fold_2/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 36, 37, 38, 39, 40, 41, 43, 44, 45, 47, 48, 49, 50, 51, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 68, 69, 70, 72, 74, 76, 78, 79, 80, 82, 83, 84, 85, 86, 87, 88, 91, 93, 94, 95, 96, 97, 100, 101, 102, 103, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 142, 143, 145, 147, 148, 149, 150, 151, 152, 153, 156, 157, 158, 159, 160, 161, 162, 163, 165, 166, 167, 168, 170, 171, 172, 173, 174, 175, 177, 179, 180, 181, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 196, 197, 199, 200, 202, 203, 204, 205, 207, 208, 209, 210, 211, 212, 213, 214, 215, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 232, 233, 235, 236, 237, 238, 239, 240, 241, 242, 243, 245, 246, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 264, 265, 266, 267, 268, 269, 271, 272, 273, 274, 275, 276, 280, 281, 282, 283, 284, 285, 287, 289, 290, 293, 294, 295, 296, 297, 298, 299, 300, 302, 303, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 317, 318, 320, 321, 322, 323, 324, 326, 327, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 358, 360, 361, 363, 364, 365, 366, 368, 369, 370, 371, 372, 373, 376, 377, 378, 379, 380, 381, 382, 383, 385, 386, 387, 388, 389, 390, 392, 394, 395, 398, 399, 400, 401, 402, 404, 405, 407, 408, 409, 410, 412, 413, 414, 415, 416, 417, 418], "outer_test_idx": [10, 12, 15, 31, 35, 42, 46, 52, 53, 54, 62, 66, 67, 71, 73, 75, 77, 81, 89, 90, 92, 98, 99, 105, 123, 124, 141, 144, 146, 154, 155, 164, 169, 176, 178, 182, 189, 190, 195, 198, 201, 206, 216, 217, 218, 219, 231, 234, 244, 247, 248, 258, 262, 270, 277, 278, 279, 286, 288, 291, 292, 301, 304, 306, 316, 319, 325, 328, 343, 357, 359, 362, 367, 374, 375, 384, 391, 393, 396, 397, 403, 406, 411, 419]}

42
models/abl/both/20260617_232841/outer_fold_2/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.13105160735348353,
"rmse": 0.36201050724182515,
"mae": 0.29236553396497456,
"r2": -0.4041762321007929
},
"delivery": {
"n_samples": 60,
"mse": 0.6170164560338233,
"rmse": 0.7855039503616918,
"mae": 0.518595163594,
"r2": 0.178644153319804
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5714285714285714,
"precision": 0.3343023255813954,
"recall": 0.5261136712749617,
"f1": 0.34470899470899463
},
"ee": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.6311332192269291,
"recall": 0.6916433239962653,
"f1": 0.6470105449067581
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.2793246003181208,
"js_divergence": 0.07360589998440034
}
}

16
models/abl/both/20260617_232841/outer_fold_3/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.11061601330629428,
"lr": 0.0003491022701087641,
"weight_decay": 1.2696904821623371e-05,
"backbone_lr_ratio": 0.39277239063286945,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/outer_fold_3/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 4, 5, 7, 8, 9, 10, 11, 12, 14, 15, 16, 17, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 45, 46, 47, 51, 52, 53, 54, 55, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 79, 81, 82, 83, 84, 85, 86, 87, 89, 90, 91, 92, 93, 95, 96, 97, 98, 99, 100, 102, 104, 105, 106, 109, 110, 111, 112, 113, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 127, 129, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 146, 147, 148, 150, 151, 152, 153, 154, 155, 156, 158, 159, 160, 161, 162, 163, 164, 165, 166, 168, 169, 170, 172, 173, 174, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 189, 190, 191, 192, 193, 194, 195, 198, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 212, 213, 214, 215, 216, 217, 218, 219, 221, 223, 224, 226, 227, 228, 230, 231, 233, 234, 235, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 251, 252, 253, 254, 255, 256, 257, 258, 259, 262, 263, 264, 265, 266, 267, 268, 269, 270, 272, 273, 274, 275, 276, 277, 278, 279, 280, 282, 284, 286, 288, 289, 290, 291, 292, 293, 294, 295, 297, 298, 299, 301, 302, 304, 305, 306, 307, 308, 309, 310, 312, 314, 315, 316, 317, 319, 320, 321, 325, 326, 327, 328, 331, 332, 334, 336, 340, 341, 343, 344, 346, 347, 348, 350, 352, 353, 357, 359, 360, 361, 362, 363, 364, 366, 367, 368, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 382, 383, 384, 385, 386, 387, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 400, 401, 402, 403, 404, 406, 407, 410, 411, 412, 414, 415, 416, 418, 419], "outer_test_idx": [3, 6, 13, 18, 19, 26, 33, 44, 48, 49, 50, 56, 57, 72, 80, 88, 94, 101, 103, 107, 108, 114, 126, 128, 130, 145, 149, 157, 167, 171, 175, 188, 196, 197, 199, 211, 220, 222, 225, 229, 232, 236, 250, 260, 261, 271, 281, 283, 285, 287, 296, 300, 303, 311, 313, 318, 322, 323, 324, 329, 330, 333, 335, 337, 338, 339, 342, 345, 349, 351, 354, 355, 356, 358, 365, 369, 381, 388, 399, 405, 408, 409, 413, 417]}

42
models/abl/both/20260617_232841/outer_fold_3/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.2974639059494916,
"rmse": 0.5454025173662949,
"mae": 0.2553033943635872,
"r2": 0.08722103632340716
},
"delivery": {
"n_samples": 59,
"mse": 0.72832486531675,
"rmse": 0.8534195130864715,
"mae": 0.5703363718970095,
"r2": 0.29198169797506535
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.3663025210084033,
"recall": 0.5624039938556068,
"f1": 0.3811802616390919
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.5903252773977808,
"recall": 0.6201608848667672,
"f1": 0.59879647539222
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.4182398191784831,
"js_divergence": 0.10764099439445415
}
}

16
models/abl/both/20260617_232841/outer_fold_4/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.41353294032230026,
"lr": 0.00018606616740103076,
"weight_decay": 9.912252717813433e-05,
"backbone_lr_ratio": 0.32820993966799533,
"moe_n_experts": 4,
"moe_top_k": 1,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/outer_fold_4/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 23, 25, 26, 28, 29, 31, 33, 34, 35, 37, 41, 42, 43, 44, 45, 46, 48, 49, 50, 51, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 107, 108, 110, 111, 112, 113, 114, 116, 117, 119, 120, 121, 122, 123, 124, 126, 127, 128, 130, 132, 133, 134, 135, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 149, 152, 154, 155, 157, 160, 161, 162, 163, 164, 165, 166, 167, 169, 170, 171, 173, 174, 175, 176, 177, 178, 180, 181, 182, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 205, 206, 207, 208, 209, 211, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 229, 230, 231, 232, 234, 235, 236, 238, 239, 240, 241, 243, 244, 245, 246, 247, 248, 249, 250, 251, 253, 254, 257, 258, 260, 261, 262, 263, 264, 265, 266, 267, 270, 271, 272, 273, 277, 278, 279, 280, 281, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 296, 298, 299, 300, 301, 302, 303, 304, 305, 306, 309, 311, 312, 313, 314, 315, 316, 318, 319, 320, 322, 323, 324, 325, 326, 328, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 341, 342, 343, 345, 346, 347, 349, 350, 351, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 365, 367, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 381, 383, 384, 385, 386, 388, 390, 391, 392, 393, 394, 395, 396, 397, 399, 400, 401, 402, 403, 405, 406, 407, 408, 409, 410, 411, 413, 415, 416, 417, 418, 419], "outer_test_idx": [0, 5, 11, 21, 24, 27, 30, 32, 36, 38, 39, 40, 47, 55, 58, 61, 65, 76, 86, 95, 100, 106, 109, 115, 118, 125, 129, 131, 136, 148, 150, 151, 153, 156, 158, 159, 168, 172, 179, 183, 184, 185, 202, 204, 210, 212, 213, 228, 233, 237, 242, 252, 255, 256, 259, 268, 269, 274, 275, 276, 282, 297, 307, 308, 310, 317, 321, 327, 340, 344, 348, 352, 364, 366, 368, 379, 380, 382, 387, 389, 398, 404, 412, 414]}

42
models/abl/both/20260617_232841/outer_fold_4/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.2385472153185827,
"rmse": 0.4884129557235175,
"mae": 0.37417884383882793,
"r2": -0.7317567503291509
},
"delivery": {
"n_samples": 58,
"mse": 0.7980907160636062,
"rmse": 0.8933592312522473,
"mae": 0.6033944039252298,
"r2": 0.2282594525293985
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5952380952380952,
"precision": 0.33809196980096085,
"recall": 0.5263056835637481,
"f1": 0.3583862620559868
},
"ee": {
"n_samples": 84,
"accuracy": 0.7261904761904762,
"precision": 0.6831659493065518,
"recall": 0.7594017094017094,
"f1": 0.7065123352302344
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.4050166777635646,
"js_divergence": 0.10143992174009567
}
}

16
models/abl/both/20260617_232841/repeat_1/outer_fold_0/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.12930599750984867,
"lr": 0.0003183893957684731,
"weight_decay": 0.0011797820449629415,
"backbone_lr_ratio": 0.35598913444921026,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_1/outer_fold_0/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72, 73, 75, 76, 77, 80, 81, 85, 86, 88, 89, 90, 92, 94, 95, 96, 98, 99, 100, 101, 103, 104, 105, 106, 107, 108, 109, 111, 113, 114, 115, 116, 117, 118, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 133, 134, 135, 136, 138, 139, 141, 144, 145, 146, 147, 148, 149, 150, 151, 153, 154, 155, 156, 157, 158, 159, 160, 163, 164, 167, 168, 169, 170, 171, 172, 175, 176, 178, 179, 180, 182, 183, 184, 185, 188, 189, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 206, 208, 210, 211, 212, 213, 216, 217, 218, 219, 220, 222, 224, 225, 227, 228, 229, 230, 231, 232, 233, 234, 236, 237, 239, 241, 242, 244, 245, 247, 248, 249, 250, 252, 253, 255, 256, 257, 258, 259, 260, 261, 262, 265, 267, 268, 269, 270, 271, 272, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 285, 286, 287, 288, 291, 292, 294, 296, 297, 298, 299, 300, 301, 303, 304, 306, 307, 308, 309, 310, 311, 312, 313, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 333, 335, 336, 337, 338, 339, 340, 342, 343, 344, 345, 346, 347, 348, 349, 351, 352, 354, 355, 356, 357, 358, 359, 360, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372, 374, 375, 377, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 393, 395, 396, 397, 398, 399, 400, 401, 403, 404, 405, 406, 407, 408, 409, 411, 412, 413, 414, 415, 417, 418, 419], "outer_test_idx": [7, 16, 20, 29, 51, 59, 60, 69, 70, 74, 78, 79, 82, 83, 84, 87, 91, 93, 97, 102, 110, 112, 119, 120, 121, 132, 137, 140, 142, 143, 152, 161, 162, 165, 166, 173, 174, 177, 181, 186, 187, 194, 205, 207, 209, 214, 215, 221, 223, 226, 235, 238, 240, 243, 246, 251, 254, 263, 264, 266, 273, 284, 289, 290, 293, 295, 302, 305, 314, 315, 331, 334, 341, 350, 353, 361, 373, 376, 378, 392, 394, 402, 410, 416]}

42
models/abl/both/20260617_232841/repeat_1/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.2816563970171338,
"rmse": 0.5307131023605257,
"mae": 0.27628492734518395,
"r2": 0.06479970039267835
},
"delivery": {
"n_samples": 58,
"mse": 0.5807003895647976,
"rmse": 0.7620370001284699,
"mae": 0.564853833378132,
"r2": 0.26189476312686466
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7857142857142857,
"precision": 0.42781279178338005,
"recall": 0.5944574551131928,
"f1": 0.4533531409168081
},
"ee": {
"n_samples": 84,
"accuracy": 0.6547619047619048,
"precision": 0.6603426956368134,
"recall": 0.7080952380952382,
"f1": 0.6472175061309188
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9827586206896551,
"precision": 0.9912280701754386,
"recall": 0.75,
"f1": 0.8289085545722714
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.2790682434585921,
"js_divergence": 0.06757546969442413
}
}

16
models/abl/both/20260617_232841/repeat_1/outer_fold_1/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.14959924395216218,
"lr": 0.00012401964324838103,
"weight_decay": 9.656428689903883e-05,
"backbone_lr_ratio": 0.32683834688760083,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_1/outer_fold_1/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 3, 5, 6, 7, 10, 11, 12, 13, 15, 16, 18, 19, 20, 21, 24, 26, 27, 29, 30, 31, 32, 33, 35, 36, 38, 39, 40, 42, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 97, 98, 99, 100, 101, 102, 103, 105, 106, 107, 108, 109, 110, 112, 114, 115, 118, 119, 120, 121, 123, 124, 125, 126, 128, 129, 130, 131, 132, 136, 137, 140, 141, 142, 143, 144, 145, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 161, 162, 164, 165, 166, 167, 168, 169, 171, 172, 173, 174, 175, 176, 177, 178, 179, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 194, 195, 196, 197, 198, 199, 201, 202, 204, 205, 206, 207, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 244, 246, 247, 248, 250, 251, 252, 254, 255, 256, 258, 259, 260, 261, 262, 263, 264, 266, 268, 269, 270, 271, 273, 274, 275, 276, 277, 278, 279, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 300, 301, 302, 303, 304, 305, 306, 307, 308, 310, 311, 313, 314, 315, 316, 317, 318, 319, 321, 322, 323, 324, 325, 327, 328, 329, 330, 331, 333, 334, 335, 337, 338, 339, 340, 341, 342, 343, 344, 345, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 361, 362, 364, 365, 366, 367, 368, 369, 373, 374, 375, 376, 378, 379, 380, 381, 382, 384, 387, 388, 389, 391, 392, 393, 394, 396, 397, 398, 399, 402, 403, 404, 405, 406, 408, 409, 410, 411, 412, 413, 414, 416, 417, 419], "outer_test_idx": [1, 2, 4, 8, 9, 14, 17, 22, 23, 25, 28, 34, 37, 41, 43, 45, 63, 64, 68, 85, 96, 104, 111, 113, 116, 117, 122, 127, 133, 134, 135, 138, 139, 147, 160, 163, 170, 180, 191, 192, 193, 200, 203, 208, 224, 227, 230, 239, 241, 245, 249, 253, 257, 265, 267, 272, 280, 294, 298, 299, 309, 312, 320, 326, 332, 336, 346, 347, 360, 363, 370, 371, 372, 377, 383, 385, 386, 390, 395, 400, 401, 407, 415, 418]}

42
models/abl/both/20260617_232841/repeat_1/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.08666696329506099,
"rmse": 0.2943925326754417,
"mae": 0.22798642090388707,
"r2": -0.4357809346276156
},
"delivery": {
"n_samples": 61,
"mse": 0.9690008398737997,
"rmse": 0.9843784027871597,
"mae": 0.626403434202075,
"r2": 0.26653621845095465
},
"pdi": {
"n_samples": 84,
"accuracy": 0.4880952380952381,
"precision": 0.27691029900332226,
"recall": 0.2579365079365079,
"f1": 0.24170918367346939
},
"ee": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.4703263535182041,
"recall": 0.47331240188383045,
"f1": 0.46348793258357346
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.4359101867730315,
"js_divergence": 0.11254492304386336
}
}

16
models/abl/both/20260617_232841/repeat_1/outer_fold_2/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.12302210383900847,
"lr": 0.00015510064533248203,
"weight_decay": 0.0037340196924722196,
"backbone_lr_ratio": 0.3036603338505521,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_1/outer_fold_2/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 36, 37, 38, 39, 40, 41, 43, 44, 45, 47, 48, 49, 50, 51, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 68, 69, 70, 72, 74, 76, 78, 79, 80, 82, 83, 84, 85, 86, 87, 88, 91, 93, 94, 95, 96, 97, 100, 101, 102, 103, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 142, 143, 145, 147, 148, 149, 150, 151, 152, 153, 156, 157, 158, 159, 160, 161, 162, 163, 165, 166, 167, 168, 170, 171, 172, 173, 174, 175, 177, 179, 180, 181, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 196, 197, 199, 200, 202, 203, 204, 205, 207, 208, 209, 210, 211, 212, 213, 214, 215, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 232, 233, 235, 236, 237, 238, 239, 240, 241, 242, 243, 245, 246, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 264, 265, 266, 267, 268, 269, 271, 272, 273, 274, 275, 276, 280, 281, 282, 283, 284, 285, 287, 289, 290, 293, 294, 295, 296, 297, 298, 299, 300, 302, 303, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 317, 318, 320, 321, 322, 323, 324, 326, 327, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 358, 360, 361, 363, 364, 365, 366, 368, 369, 370, 371, 372, 373, 376, 377, 378, 379, 380, 381, 382, 383, 385, 386, 387, 388, 389, 390, 392, 394, 395, 398, 399, 400, 401, 402, 404, 405, 407, 408, 409, 410, 412, 413, 414, 415, 416, 417, 418], "outer_test_idx": [10, 12, 15, 31, 35, 42, 46, 52, 53, 54, 62, 66, 67, 71, 73, 75, 77, 81, 89, 90, 92, 98, 99, 105, 123, 124, 141, 144, 146, 154, 155, 164, 169, 176, 178, 182, 189, 190, 195, 198, 201, 206, 216, 217, 218, 219, 231, 234, 244, 247, 248, 258, 262, 270, 277, 278, 279, 286, 288, 291, 292, 301, 304, 306, 316, 319, 325, 328, 343, 357, 359, 362, 367, 374, 375, 384, 391, 393, 396, 397, 403, 406, 411, 419]}

42
models/abl/both/20260617_232841/repeat_1/outer_fold_2/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.13747360050361937,
"rmse": 0.3707743255723343,
"mae": 0.2868303656578064,
"r2": -0.4729858432626901
},
"delivery": {
"n_samples": 60,
"mse": 0.7104536138129312,
"rmse": 0.8428841046151785,
"mae": 0.5430462850956246,
"r2": 0.054263101423121185
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6428571428571429,
"precision": 0.45827160493827157,
"recall": 0.7127496159754224,
"f1": 0.4944194569381975
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6060343270099368,
"recall": 0.6573295985060691,
"f1": 0.6198067632850242
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.28003378134681434,
"js_divergence": 0.0731604987811758
}
}

16
models/abl/both/20260617_232841/repeat_1/outer_fold_3/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.11061601330629428,
"lr": 0.0003491022701087641,
"weight_decay": 1.2696904821623371e-05,
"backbone_lr_ratio": 0.39277239063286945,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_1/outer_fold_3/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 4, 5, 7, 8, 9, 10, 11, 12, 14, 15, 16, 17, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 45, 46, 47, 51, 52, 53, 54, 55, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 79, 81, 82, 83, 84, 85, 86, 87, 89, 90, 91, 92, 93, 95, 96, 97, 98, 99, 100, 102, 104, 105, 106, 109, 110, 111, 112, 113, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 127, 129, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 146, 147, 148, 150, 151, 152, 153, 154, 155, 156, 158, 159, 160, 161, 162, 163, 164, 165, 166, 168, 169, 170, 172, 173, 174, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 189, 190, 191, 192, 193, 194, 195, 198, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 212, 213, 214, 215, 216, 217, 218, 219, 221, 223, 224, 226, 227, 228, 230, 231, 233, 234, 235, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 251, 252, 253, 254, 255, 256, 257, 258, 259, 262, 263, 264, 265, 266, 267, 268, 269, 270, 272, 273, 274, 275, 276, 277, 278, 279, 280, 282, 284, 286, 288, 289, 290, 291, 292, 293, 294, 295, 297, 298, 299, 301, 302, 304, 305, 306, 307, 308, 309, 310, 312, 314, 315, 316, 317, 319, 320, 321, 325, 326, 327, 328, 331, 332, 334, 336, 340, 341, 343, 344, 346, 347, 348, 350, 352, 353, 357, 359, 360, 361, 362, 363, 364, 366, 367, 368, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 382, 383, 384, 385, 386, 387, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 400, 401, 402, 403, 404, 406, 407, 410, 411, 412, 414, 415, 416, 418, 419], "outer_test_idx": [3, 6, 13, 18, 19, 26, 33, 44, 48, 49, 50, 56, 57, 72, 80, 88, 94, 101, 103, 107, 108, 114, 126, 128, 130, 145, 149, 157, 167, 171, 175, 188, 196, 197, 199, 211, 220, 222, 225, 229, 232, 236, 250, 260, 261, 271, 281, 283, 285, 287, 296, 300, 303, 311, 313, 318, 322, 323, 324, 329, 330, 333, 335, 337, 338, 339, 342, 345, 349, 351, 354, 355, 356, 358, 365, 369, 381, 388, 399, 405, 408, 409, 413, 417]}

42
models/abl/both/20260617_232841/repeat_1/outer_fold_3/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.32669708395989955,
"rmse": 0.5715742156184965,
"mae": 0.31596550309514426,
"r2": -0.0024820483050997932
},
"delivery": {
"n_samples": 59,
"mse": 0.7029380513850005,
"rmse": 0.8384140095352657,
"mae": 0.5650059677881457,
"r2": 0.31666069734708746
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.3406323185011709,
"recall": 0.5230414746543779,
"f1": 0.3490853658536585
},
"ee": {
"n_samples": 84,
"accuracy": 0.6547619047619048,
"precision": 0.5888888888888889,
"recall": 0.6327300150829562,
"f1": 0.5970695970695971
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3455956570999495,
"js_divergence": 0.08991455654751294
}
}

16
models/abl/both/20260617_232841/repeat_1/outer_fold_4/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.41353294032230026,
"lr": 0.00018606616740103076,
"weight_decay": 9.912252717813433e-05,
"backbone_lr_ratio": 0.32820993966799533,
"moe_n_experts": 4,
"moe_top_k": 1,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_1/outer_fold_4/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 23, 25, 26, 28, 29, 31, 33, 34, 35, 37, 41, 42, 43, 44, 45, 46, 48, 49, 50, 51, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 107, 108, 110, 111, 112, 113, 114, 116, 117, 119, 120, 121, 122, 123, 124, 126, 127, 128, 130, 132, 133, 134, 135, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 149, 152, 154, 155, 157, 160, 161, 162, 163, 164, 165, 166, 167, 169, 170, 171, 173, 174, 175, 176, 177, 178, 180, 181, 182, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 205, 206, 207, 208, 209, 211, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 229, 230, 231, 232, 234, 235, 236, 238, 239, 240, 241, 243, 244, 245, 246, 247, 248, 249, 250, 251, 253, 254, 257, 258, 260, 261, 262, 263, 264, 265, 266, 267, 270, 271, 272, 273, 277, 278, 279, 280, 281, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 296, 298, 299, 300, 301, 302, 303, 304, 305, 306, 309, 311, 312, 313, 314, 315, 316, 318, 319, 320, 322, 323, 324, 325, 326, 328, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 341, 342, 343, 345, 346, 347, 349, 350, 351, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 365, 367, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 381, 383, 384, 385, 386, 388, 390, 391, 392, 393, 394, 395, 396, 397, 399, 400, 401, 402, 403, 405, 406, 407, 408, 409, 410, 411, 413, 415, 416, 417, 418, 419], "outer_test_idx": [0, 5, 11, 21, 24, 27, 30, 32, 36, 38, 39, 40, 47, 55, 58, 61, 65, 76, 86, 95, 100, 106, 109, 115, 118, 125, 129, 131, 136, 148, 150, 151, 153, 156, 158, 159, 168, 172, 179, 183, 184, 185, 202, 204, 210, 212, 213, 228, 233, 237, 242, 252, 255, 256, 259, 268, 269, 274, 275, 276, 282, 297, 307, 308, 310, 317, 321, 327, 340, 344, 348, 352, 364, 366, 368, 379, 380, 382, 387, 389, 398, 404, 412, 414]}

42
models/abl/both/20260617_232841/repeat_1/outer_fold_4/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.2498053256210641,
"rmse": 0.49980528770818755,
"mae": 0.4012378028460911,
"r2": -0.8134860989037498
},
"delivery": {
"n_samples": 58,
"mse": 0.8713336067211406,
"rmse": 0.9334525198000917,
"mae": 0.6369200194704121,
"r2": 0.1574347863646688
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6190476190476191,
"precision": 0.32608695652173914,
"recall": 0.510752688172043,
"f1": 0.350912975912976
},
"ee": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.6324786324786325,
"recall": 0.7021367521367522,
"f1": 0.6488452445899254
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3635802403548602,
"js_divergence": 0.09224357999051915
}
}

16
models/abl/both/20260617_232841/repeat_2/outer_fold_0/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.12930599750984867,
"lr": 0.0003183893957684731,
"weight_decay": 0.0011797820449629415,
"backbone_lr_ratio": 0.35598913444921026,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_2/outer_fold_0/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72, 73, 75, 76, 77, 80, 81, 85, 86, 88, 89, 90, 92, 94, 95, 96, 98, 99, 100, 101, 103, 104, 105, 106, 107, 108, 109, 111, 113, 114, 115, 116, 117, 118, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 133, 134, 135, 136, 138, 139, 141, 144, 145, 146, 147, 148, 149, 150, 151, 153, 154, 155, 156, 157, 158, 159, 160, 163, 164, 167, 168, 169, 170, 171, 172, 175, 176, 178, 179, 180, 182, 183, 184, 185, 188, 189, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 206, 208, 210, 211, 212, 213, 216, 217, 218, 219, 220, 222, 224, 225, 227, 228, 229, 230, 231, 232, 233, 234, 236, 237, 239, 241, 242, 244, 245, 247, 248, 249, 250, 252, 253, 255, 256, 257, 258, 259, 260, 261, 262, 265, 267, 268, 269, 270, 271, 272, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 285, 286, 287, 288, 291, 292, 294, 296, 297, 298, 299, 300, 301, 303, 304, 306, 307, 308, 309, 310, 311, 312, 313, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 333, 335, 336, 337, 338, 339, 340, 342, 343, 344, 345, 346, 347, 348, 349, 351, 352, 354, 355, 356, 357, 358, 359, 360, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372, 374, 375, 377, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 393, 395, 396, 397, 398, 399, 400, 401, 403, 404, 405, 406, 407, 408, 409, 411, 412, 413, 414, 415, 417, 418, 419], "outer_test_idx": [7, 16, 20, 29, 51, 59, 60, 69, 70, 74, 78, 79, 82, 83, 84, 87, 91, 93, 97, 102, 110, 112, 119, 120, 121, 132, 137, 140, 142, 143, 152, 161, 162, 165, 166, 173, 174, 177, 181, 186, 187, 194, 205, 207, 209, 214, 215, 221, 223, 226, 235, 238, 240, 243, 246, 251, 254, 263, 264, 266, 273, 284, 289, 290, 293, 295, 302, 305, 314, 315, 331, 334, 341, 350, 353, 361, 373, 376, 378, 392, 394, 402, 410, 416]}

42
models/abl/both/20260617_232841/repeat_2/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.3048058934964392,
"rmse": 0.5520922871191366,
"mae": 0.30447895555611115,
"r2": -0.012064934220566759
},
"delivery": {
"n_samples": 58,
"mse": 0.5277170018903071,
"rmse": 0.7264413272180398,
"mae": 0.553587457165122,
"r2": 0.32923984608630563
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7261904761904762,
"precision": 0.3958333333333333,
"recall": 0.5817720530835285,
"f1": 0.43162393162393164
},
"ee": {
"n_samples": 84,
"accuracy": 0.75,
"precision": 0.7124588290649997,
"recall": 0.7785714285714285,
"f1": 0.7248325686863893
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.2720172018326682,
"js_divergence": 0.060785283409926956
}
}

16
models/abl/both/20260617_232841/repeat_2/outer_fold_1/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.14959924395216218,
"lr": 0.00012401964324838103,
"weight_decay": 9.656428689903883e-05,
"backbone_lr_ratio": 0.32683834688760083,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_2/outer_fold_1/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 3, 5, 6, 7, 10, 11, 12, 13, 15, 16, 18, 19, 20, 21, 24, 26, 27, 29, 30, 31, 32, 33, 35, 36, 38, 39, 40, 42, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 97, 98, 99, 100, 101, 102, 103, 105, 106, 107, 108, 109, 110, 112, 114, 115, 118, 119, 120, 121, 123, 124, 125, 126, 128, 129, 130, 131, 132, 136, 137, 140, 141, 142, 143, 144, 145, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 161, 162, 164, 165, 166, 167, 168, 169, 171, 172, 173, 174, 175, 176, 177, 178, 179, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 194, 195, 196, 197, 198, 199, 201, 202, 204, 205, 206, 207, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 244, 246, 247, 248, 250, 251, 252, 254, 255, 256, 258, 259, 260, 261, 262, 263, 264, 266, 268, 269, 270, 271, 273, 274, 275, 276, 277, 278, 279, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 300, 301, 302, 303, 304, 305, 306, 307, 308, 310, 311, 313, 314, 315, 316, 317, 318, 319, 321, 322, 323, 324, 325, 327, 328, 329, 330, 331, 333, 334, 335, 337, 338, 339, 340, 341, 342, 343, 344, 345, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 361, 362, 364, 365, 366, 367, 368, 369, 373, 374, 375, 376, 378, 379, 380, 381, 382, 384, 387, 388, 389, 391, 392, 393, 394, 396, 397, 398, 399, 402, 403, 404, 405, 406, 408, 409, 410, 411, 412, 413, 414, 416, 417, 419], "outer_test_idx": [1, 2, 4, 8, 9, 14, 17, 22, 23, 25, 28, 34, 37, 41, 43, 45, 63, 64, 68, 85, 96, 104, 111, 113, 116, 117, 122, 127, 133, 134, 135, 138, 139, 147, 160, 163, 170, 180, 191, 192, 193, 200, 203, 208, 224, 227, 230, 239, 241, 245, 249, 253, 257, 265, 267, 272, 280, 294, 298, 299, 309, 312, 320, 326, 332, 336, 346, 347, 360, 363, 370, 371, 372, 377, 383, 385, 386, 390, 395, 400, 401, 407, 415, 418]}

42
models/abl/both/20260617_232841/repeat_2/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.10482404388180623,
"rmse": 0.32376541489449767,
"mae": 0.23482046808515275,
"r2": -0.7365828681878246
},
"delivery": {
"n_samples": 61,
"mse": 0.9723624200125729,
"rmse": 0.9860843878758921,
"mae": 0.6302215732084434,
"r2": 0.2639917446186245
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5952380952380952,
"precision": 0.28258620689655173,
"recall": 0.2698412698412698,
"f1": 0.2714159292035398
},
"ee": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.48611111111111116,
"recall": 0.4884639170353456,
"f1": 0.4753962887837127
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.4600248432068658,
"js_divergence": 0.11821523045175072
}
}

16
models/abl/both/20260617_232841/repeat_2/outer_fold_2/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.12302210383900847,
"lr": 0.00015510064533248203,
"weight_decay": 0.0037340196924722196,
"backbone_lr_ratio": 0.3036603338505521,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_2/outer_fold_2/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 36, 37, 38, 39, 40, 41, 43, 44, 45, 47, 48, 49, 50, 51, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 68, 69, 70, 72, 74, 76, 78, 79, 80, 82, 83, 84, 85, 86, 87, 88, 91, 93, 94, 95, 96, 97, 100, 101, 102, 103, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 142, 143, 145, 147, 148, 149, 150, 151, 152, 153, 156, 157, 158, 159, 160, 161, 162, 163, 165, 166, 167, 168, 170, 171, 172, 173, 174, 175, 177, 179, 180, 181, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 196, 197, 199, 200, 202, 203, 204, 205, 207, 208, 209, 210, 211, 212, 213, 214, 215, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 232, 233, 235, 236, 237, 238, 239, 240, 241, 242, 243, 245, 246, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 264, 265, 266, 267, 268, 269, 271, 272, 273, 274, 275, 276, 280, 281, 282, 283, 284, 285, 287, 289, 290, 293, 294, 295, 296, 297, 298, 299, 300, 302, 303, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 317, 318, 320, 321, 322, 323, 324, 326, 327, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 358, 360, 361, 363, 364, 365, 366, 368, 369, 370, 371, 372, 373, 376, 377, 378, 379, 380, 381, 382, 383, 385, 386, 387, 388, 389, 390, 392, 394, 395, 398, 399, 400, 401, 402, 404, 405, 407, 408, 409, 410, 412, 413, 414, 415, 416, 417, 418], "outer_test_idx": [10, 12, 15, 31, 35, 42, 46, 52, 53, 54, 62, 66, 67, 71, 73, 75, 77, 81, 89, 90, 92, 98, 99, 105, 123, 124, 141, 144, 146, 154, 155, 164, 169, 176, 178, 182, 189, 190, 195, 198, 201, 206, 216, 217, 218, 219, 231, 234, 244, 247, 248, 258, 262, 270, 277, 278, 279, 286, 288, 291, 292, 301, 304, 306, 316, 319, 325, 328, 343, 357, 359, 362, 367, 374, 375, 384, 391, 393, 396, 397, 403, 406, 411, 419]}

42
models/abl/both/20260617_232841/repeat_2/outer_fold_2/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.12549337649644715,
"rmse": 0.35425044318454585,
"mae": 0.2802231794311887,
"r2": -0.34462155879619094
},
"delivery": {
"n_samples": 60,
"mse": 0.6592902402854742,
"rmse": 0.8119668960527111,
"mae": 0.5348013029200956,
"r2": 0.12237041942365745
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5119047619047619,
"precision": 0.4135659661975451,
"recall": 0.6746031746031745,
"f1": 0.4122222222222222
},
"ee": {
"n_samples": 84,
"accuracy": 0.6547619047619048,
"precision": 0.5972649572649572,
"recall": 0.6507936507936507,
"f1": 0.6069243156199677
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.2563386788633516,
"js_divergence": 0.06872839042316117
}
}

16
models/abl/both/20260617_232841/repeat_2/outer_fold_3/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.11061601330629428,
"lr": 0.0003491022701087641,
"weight_decay": 1.2696904821623371e-05,
"backbone_lr_ratio": 0.39277239063286945,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_2/outer_fold_3/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 4, 5, 7, 8, 9, 10, 11, 12, 14, 15, 16, 17, 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 45, 46, 47, 51, 52, 53, 54, 55, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 79, 81, 82, 83, 84, 85, 86, 87, 89, 90, 91, 92, 93, 95, 96, 97, 98, 99, 100, 102, 104, 105, 106, 109, 110, 111, 112, 113, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 127, 129, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 146, 147, 148, 150, 151, 152, 153, 154, 155, 156, 158, 159, 160, 161, 162, 163, 164, 165, 166, 168, 169, 170, 172, 173, 174, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 189, 190, 191, 192, 193, 194, 195, 198, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 212, 213, 214, 215, 216, 217, 218, 219, 221, 223, 224, 226, 227, 228, 230, 231, 233, 234, 235, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 251, 252, 253, 254, 255, 256, 257, 258, 259, 262, 263, 264, 265, 266, 267, 268, 269, 270, 272, 273, 274, 275, 276, 277, 278, 279, 280, 282, 284, 286, 288, 289, 290, 291, 292, 293, 294, 295, 297, 298, 299, 301, 302, 304, 305, 306, 307, 308, 309, 310, 312, 314, 315, 316, 317, 319, 320, 321, 325, 326, 327, 328, 331, 332, 334, 336, 340, 341, 343, 344, 346, 347, 348, 350, 352, 353, 357, 359, 360, 361, 362, 363, 364, 366, 367, 368, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 382, 383, 384, 385, 386, 387, 389, 390, 391, 392, 393, 394, 395, 396, 397, 398, 400, 401, 402, 403, 404, 406, 407, 410, 411, 412, 414, 415, 416, 418, 419], "outer_test_idx": [3, 6, 13, 18, 19, 26, 33, 44, 48, 49, 50, 56, 57, 72, 80, 88, 94, 101, 103, 107, 108, 114, 126, 128, 130, 145, 149, 157, 167, 171, 175, 188, 196, 197, 199, 211, 220, 222, 225, 229, 232, 236, 250, 260, 261, 271, 281, 283, 285, 287, 296, 300, 303, 311, 313, 318, 322, 323, 324, 329, 330, 333, 335, 337, 338, 339, 342, 345, 349, 351, 354, 355, 356, 358, 365, 369, 381, 388, 399, 405, 408, 409, 413, 417]}

42
models/abl/both/20260617_232841/repeat_2/outer_fold_3/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.30582211903190704,
"rmse": 0.5530118615652896,
"mae": 0.297713055668107,
"r2": 0.06157355129088471
},
"delivery": {
"n_samples": 59,
"mse": 0.7117105624010245,
"rmse": 0.8436293987296937,
"mae": 0.5593153651077616,
"r2": 0.30813277436953124
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.36681547619047616,
"recall": 0.5662442396313364,
"f1": 0.3804035250463822
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.5864396721189867,
"recall": 0.6201608848667672,
"f1": 0.5949003714961162
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.31884722326826903,
"js_divergence": 0.08589330459529161
}
}

16
models/abl/both/20260617_232841/repeat_2/outer_fold_4/best_params.json
View File

@ -1,16 +0,0 @@
{
"dropout": 0.41353294032230026,
"lr": 0.00018606616740103076,
"weight_decay": 9.912252717813433e-05,
"backbone_lr_ratio": 0.32820993966799533,
"moe_n_experts": 4,
"moe_top_k": 1,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
}

1
models/abl/both/20260617_232841/repeat_2/outer_fold_4/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 23, 25, 26, 28, 29, 31, 33, 34, 35, 37, 41, 42, 43, 44, 45, 46, 48, 49, 50, 51, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82, 83, 84, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 107, 108, 110, 111, 112, 113, 114, 116, 117, 119, 120, 121, 122, 123, 124, 126, 127, 128, 130, 132, 133, 134, 135, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 149, 152, 154, 155, 157, 160, 161, 162, 163, 164, 165, 166, 167, 169, 170, 171, 173, 174, 175, 176, 177, 178, 180, 181, 182, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 203, 205, 206, 207, 208, 209, 211, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 229, 230, 231, 232, 234, 235, 236, 238, 239, 240, 241, 243, 244, 245, 246, 247, 248, 249, 250, 251, 253, 254, 257, 258, 260, 261, 262, 263, 264, 265, 266, 267, 270, 271, 272, 273, 277, 278, 279, 280, 281, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 296, 298, 299, 300, 301, 302, 303, 304, 305, 306, 309, 311, 312, 313, 314, 315, 316, 318, 319, 320, 322, 323, 324, 325, 326, 328, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 341, 342, 343, 345, 346, 347, 349, 350, 351, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 365, 367, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 381, 383, 384, 385, 386, 388, 390, 391, 392, 393, 394, 395, 396, 397, 399, 400, 401, 402, 403, 405, 406, 407, 408, 409, 410, 411, 413, 415, 416, 417, 418, 419], "outer_test_idx": [0, 5, 11, 21, 24, 27, 30, 32, 36, 38, 39, 40, 47, 55, 58, 61, 65, 76, 86, 95, 100, 106, 109, 115, 118, 125, 129, 131, 136, 148, 150, 151, 153, 156, 158, 159, 168, 172, 179, 183, 184, 185, 202, 204, 210, 212, 213, 228, 233, 237, 242, 252, 255, 256, 259, 268, 269, 274, 275, 276, 282, 297, 307, 308, 310, 317, 321, 327, 340, 344, 348, 352, 364, 366, 368, 379, 380, 382, 387, 389, 398, 404, 412, 414]}

42
models/abl/both/20260617_232841/repeat_2/outer_fold_4/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.21199504587721935,
"rmse": 0.4604291974638656,
"mae": 0.33512775670914424,
"r2": -0.5389986893953644
},
"delivery": {
"n_samples": 58,
"mse": 0.8018784776714993,
"rmse": 0.8954766762297605,
"mae": 0.6036076779509413,
"r2": 0.22459674958332065
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6190476190476191,
"precision": 0.4494096489170381,
"recall": 0.7124935995903737,
"f1": 0.47481767481767473
},
"ee": {
"n_samples": 84,
"accuracy": 0.7142857142857143,
"precision": 0.664516711833785,
"recall": 0.7427350427350428,
"f1": 0.6874551971326165
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3693123344696232,
"js_divergence": 0.09443444364888388
}
}

60
models/abl/both/20260617_232841/strata_info.json
View File

@ -1,60 +0,0 @@
{
"original_strata_counts": {
"T0|P0|E0": "5",
"T0|P0|E1": "58",
"T0|P0|E2": "169",
"T0|P1|E0": "1",
"T0|P1|E1": "17",
"T0|P1|E2": "32",
"T0|P2|E2": "3",
"T1|P0|E2": "9",
"T1|P1|E2": "5",
"TNA|P0|E0": "28",
"TNA|P0|E1": "21",
"TNA|P0|E2": "20",
"TNA|P1|E0": "29",
"TNA|P1|E1": "7",
"TNA|P1|E2": "14",
"TNA|P2|E2": "1",
"TNA|P3|E0": "1"
},
"rare_strata": [
"T0|P1|E0",
"T0|P2|E2",
"TNA|P2|E2",
"TNA|P3|E0"
],
"final_strata": [
"RARE",
"T0|P0|E0",
"T0|P0|E1",
"T0|P0|E2",
"T0|P1|E1",
"T0|P1|E2",
"T1|P0|E2",
"T1|P1|E2",
"TNA|P0|E0",
"TNA|P0|E1",
"TNA|P0|E2",
"TNA|P1|E0",
"TNA|P1|E1",
"TNA|P1|E2"
],
"final_strata_counts": {
"RARE": "6",
"T0|P0|E0": "5",
"T0|P0|E1": "58",
"T0|P0|E2": "169",
"T0|P1|E1": "17",
"T0|P1|E2": "32",
"T1|P0|E2": "9",
"T1|P1|E2": "5",
"TNA|P0|E0": "28",
"TNA|P0|E1": "21",
"TNA|P0|E2": "20",
"TNA|P1|E0": "29",
"TNA|P1|E1": "7",
"TNA|P1|E2": "14"
},
"n_rare_merged": "6"
}

992
models/abl/both/20260617_232841/summary.json
View File

@ -1,992 +0,0 @@
{
"fold_results": [
{
"fold": 0,
"best_params": {
"dropout": 0.12930599750984867,
"lr": 0.0003183893957684731,
"weight_decay": 0.0011797820449629415,
"backbone_lr_ratio": 0.35598913444921026,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 20,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.27316252507783767,
"rmse": 0.5226495241343262,
"mae": 0.268634506018765,
"r2": 0.09300240292875017
},
"delivery": {
"n_samples": 58,
"mse": 0.49318943105662066,
"rmse": 0.7022744698881063,
"mae": 0.5225495728182381,
"r2": 0.37312647214479966
},
"pdi": {
"n_samples": 84,
"accuracy": 0.75,
"precision": 0.4107142857142857,
"recall": 0.5977751756440282,
"f1": 0.447008547008547
},
"ee": {
"n_samples": 84,
"accuracy": 0.7023809523809523,
"precision": 0.6658119658119658,
"recall": 0.6976190476190475,
"f1": 0.6690688363683327
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.20575334902695963,
"js_divergence": 0.045069148869594025
}
}
},
{
"fold": 1,
"best_params": {
"dropout": 0.14959924395216218,
"lr": 0.00012401964324838103,
"weight_decay": 9.656428689903883e-05,
"backbone_lr_ratio": 0.32683834688760083,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 23,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.11199287760383045,
"rmse": 0.3346533693298641,
"mae": 0.24370252518426805,
"r2": -0.8553464014912291
},
"delivery": {
"n_samples": 61,
"mse": 0.9945337433308287,
"rmse": 0.9972631264269368,
"mae": 0.637184505488296,
"r2": 0.24720965117371863
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5833333333333334,
"precision": 0.2606719367588933,
"recall": 0.24206349206349206,
"f1": 0.25038520801232667
},
"ee": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.4737424924924925,
"recall": 0.47331240188383045,
"f1": 0.46359870390878144
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.42904510824573017,
"js_divergence": 0.11140172525390164
}
}
},
{
"fold": 2,
"best_params": {
"dropout": 0.12302210383900847,
"lr": 0.00015510064533248203,
"weight_decay": 0.0037340196924722196,
"backbone_lr_ratio": 0.3036603338505521,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 22,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.13105160735348353,
"rmse": 0.36201050724182515,
"mae": 0.29236553396497456,
"r2": -0.4041762321007929
},
"delivery": {
"n_samples": 60,
"mse": 0.6170164560338233,
"rmse": 0.7855039503616918,
"mae": 0.518595163594,
"r2": 0.178644153319804
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5714285714285714,
"precision": 0.3343023255813954,
"recall": 0.5261136712749617,
"f1": 0.34470899470899463
},
"ee": {
"n_samples": 84,
"accuracy": 0.6904761904761905,
"precision": 0.6311332192269291,
"recall": 0.6916433239962653,
"f1": 0.6470105449067581
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.2793246003181208,
"js_divergence": 0.07360589998440034
}
}
},
{
"fold": 3,
"best_params": {
"dropout": 0.11061601330629428,
"lr": 0.0003491022701087641,
"weight_decay": 1.2696904821623371e-05,
"backbone_lr_ratio": 0.39277239063286945,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 15,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.2974639059494916,
"rmse": 0.5454025173662949,
"mae": 0.2553033943635872,
"r2": 0.08722103632340716
},
"delivery": {
"n_samples": 59,
"mse": 0.72832486531675,
"rmse": 0.8534195130864715,
"mae": 0.5703363718970095,
"r2": 0.29198169797506535
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.3663025210084033,
"recall": 0.5624039938556068,
"f1": 0.3811802616390919
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.5903252773977808,
"recall": 0.6201608848667672,
"f1": 0.59879647539222
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.4182398191784831,
"js_divergence": 0.10764099439445415
}
}
},
{
"fold": 4,
"best_params": {
"dropout": 0.41353294032230026,
"lr": 0.00018606616740103076,
"weight_decay": 9.912252717813433e-05,
"backbone_lr_ratio": 0.32820993966799533,
"moe_n_experts": 4,
"moe_top_k": 1,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 25,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.2385472153185827,
"rmse": 0.4884129557235175,
"mae": 0.37417884383882793,
"r2": -0.7317567503291509
},
"delivery": {
"n_samples": 58,
"mse": 0.7980907160636062,
"rmse": 0.8933592312522473,
"mae": 0.6033944039252298,
"r2": 0.2282594525293985
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5952380952380952,
"precision": 0.33809196980096085,
"recall": 0.5263056835637481,
"f1": 0.3583862620559868
},
"ee": {
"n_samples": 84,
"accuracy": 0.7261904761904762,
"precision": 0.6831659493065518,
"recall": 0.7594017094017094,
"f1": 0.7065123352302344
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.4050166777635646,
"js_divergence": 0.10143992174009567
}
}
},
{
"fold": 0,
"best_params": {
"dropout": 0.12930599750984867,
"lr": 0.0003183893957684731,
"weight_decay": 0.0011797820449629415,
"backbone_lr_ratio": 0.35598913444921026,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 20,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.2816563970171338,
"rmse": 0.5307131023605257,
"mae": 0.27628492734518395,
"r2": 0.06479970039267835
},
"delivery": {
"n_samples": 58,
"mse": 0.5807003895647976,
"rmse": 0.7620370001284699,
"mae": 0.564853833378132,
"r2": 0.26189476312686466
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7857142857142857,
"precision": 0.42781279178338005,
"recall": 0.5944574551131928,
"f1": 0.4533531409168081
},
"ee": {
"n_samples": 84,
"accuracy": 0.6547619047619048,
"precision": 0.6603426956368134,
"recall": 0.7080952380952382,
"f1": 0.6472175061309188
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9827586206896551,
"precision": 0.9912280701754386,
"recall": 0.75,
"f1": 0.8289085545722714
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.2790682434585921,
"js_divergence": 0.06757546969442413
}
}
},
{
"fold": 1,
"best_params": {
"dropout": 0.14959924395216218,
"lr": 0.00012401964324838103,
"weight_decay": 9.656428689903883e-05,
"backbone_lr_ratio": 0.32683834688760083,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 23,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.08666696329506099,
"rmse": 0.2943925326754417,
"mae": 0.22798642090388707,
"r2": -0.4357809346276156
},
"delivery": {
"n_samples": 61,
"mse": 0.9690008398737997,
"rmse": 0.9843784027871597,
"mae": 0.626403434202075,
"r2": 0.26653621845095465
},
"pdi": {
"n_samples": 84,
"accuracy": 0.4880952380952381,
"precision": 0.27691029900332226,
"recall": 0.2579365079365079,
"f1": 0.24170918367346939
},
"ee": {
"n_samples": 84,
"accuracy": 0.5357142857142857,
"precision": 0.4703263535182041,
"recall": 0.47331240188383045,
"f1": 0.46348793258357346
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.4359101867730315,
"js_divergence": 0.11254492304386336
}
}
},
{
"fold": 2,
"best_params": {
"dropout": 0.12302210383900847,
"lr": 0.00015510064533248203,
"weight_decay": 0.0037340196924722196,
"backbone_lr_ratio": 0.3036603338505521,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 22,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.13747360050361937,
"rmse": 0.3707743255723343,
"mae": 0.2868303656578064,
"r2": -0.4729858432626901
},
"delivery": {
"n_samples": 60,
"mse": 0.7104536138129312,
"rmse": 0.8428841046151785,
"mae": 0.5430462850956246,
"r2": 0.054263101423121185
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6428571428571429,
"precision": 0.45827160493827157,
"recall": 0.7127496159754224,
"f1": 0.4944194569381975
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.6060343270099368,
"recall": 0.6573295985060691,
"f1": 0.6198067632850242
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.28003378134681434,
"js_divergence": 0.0731604987811758
}
}
},
{
"fold": 3,
"best_params": {
"dropout": 0.11061601330629428,
"lr": 0.0003491022701087641,
"weight_decay": 1.2696904821623371e-05,
"backbone_lr_ratio": 0.39277239063286945,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 15,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.32669708395989955,
"rmse": 0.5715742156184965,
"mae": 0.31596550309514426,
"r2": -0.0024820483050997932
},
"delivery": {
"n_samples": 59,
"mse": 0.7029380513850005,
"rmse": 0.8384140095352657,
"mae": 0.5650059677881457,
"r2": 0.31666069734708746
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.3406323185011709,
"recall": 0.5230414746543779,
"f1": 0.3490853658536585
},
"ee": {
"n_samples": 84,
"accuracy": 0.6547619047619048,
"precision": 0.5888888888888889,
"recall": 0.6327300150829562,
"f1": 0.5970695970695971
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.3455956570999495,
"js_divergence": 0.08991455654751294
}
}
},
{
"fold": 4,
"best_params": {
"dropout": 0.41353294032230026,
"lr": 0.00018606616740103076,
"weight_decay": 9.912252717813433e-05,
"backbone_lr_ratio": 0.32820993966799533,
"moe_n_experts": 4,
"moe_top_k": 1,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 25,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.2498053256210641,
"rmse": 0.49980528770818755,
"mae": 0.4012378028460911,
"r2": -0.8134860989037498
},
"delivery": {
"n_samples": 58,
"mse": 0.8713336067211406,
"rmse": 0.9334525198000917,
"mae": 0.6369200194704121,
"r2": 0.1574347863646688
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6190476190476191,
"precision": 0.32608695652173914,
"recall": 0.510752688172043,
"f1": 0.350912975912976
},
"ee": {
"n_samples": 84,
"accuracy": 0.6785714285714286,
"precision": 0.6324786324786325,
"recall": 0.7021367521367522,
"f1": 0.6488452445899254
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3635802403548602,
"js_divergence": 0.09224357999051915
}
}
},
{
"fold": 0,
"best_params": {
"dropout": 0.12930599750984867,
"lr": 0.0003183893957684731,
"weight_decay": 0.0011797820449629415,
"backbone_lr_ratio": 0.35598913444921026,
"moe_n_experts": 2,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 8,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 20,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.3048058934964392,
"rmse": 0.5520922871191366,
"mae": 0.30447895555611115,
"r2": -0.012064934220566759
},
"delivery": {
"n_samples": 58,
"mse": 0.5277170018903071,
"rmse": 0.7264413272180398,
"mae": 0.553587457165122,
"r2": 0.32923984608630563
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7261904761904762,
"precision": 0.3958333333333333,
"recall": 0.5817720530835285,
"f1": 0.43162393162393164
},
"ee": {
"n_samples": 84,
"accuracy": 0.75,
"precision": 0.7124588290649997,
"recall": 0.7785714285714285,
"f1": 0.7248325686863893
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.2720172018326682,
"js_divergence": 0.060785283409926956
}
}
},
{
"fold": 1,
"best_params": {
"dropout": 0.14959924395216218,
"lr": 0.00012401964324838103,
"weight_decay": 9.656428689903883e-05,
"backbone_lr_ratio": 0.32683834688760083,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 23,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.10482404388180623,
"rmse": 0.32376541489449767,
"mae": 0.23482046808515275,
"r2": -0.7365828681878246
},
"delivery": {
"n_samples": 61,
"mse": 0.9723624200125729,
"rmse": 0.9860843878758921,
"mae": 0.6302215732084434,
"r2": 0.2639917446186245
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5952380952380952,
"precision": 0.28258620689655173,
"recall": 0.2698412698412698,
"f1": 0.2714159292035398
},
"ee": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.48611111111111116,
"recall": 0.4884639170353456,
"f1": 0.4753962887837127
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.4600248432068658,
"js_divergence": 0.11821523045175072
}
}
},
{
"fold": 2,
"best_params": {
"dropout": 0.12302210383900847,
"lr": 0.00015510064533248203,
"weight_decay": 0.0037340196924722196,
"backbone_lr_ratio": 0.3036603338505521,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 1,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 22,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.12549337649644715,
"rmse": 0.35425044318454585,
"mae": 0.2802231794311887,
"r2": -0.34462155879619094
},
"delivery": {
"n_samples": 60,
"mse": 0.6592902402854742,
"rmse": 0.8119668960527111,
"mae": 0.5348013029200956,
"r2": 0.12237041942365745
},
"pdi": {
"n_samples": 84,
"accuracy": 0.5119047619047619,
"precision": 0.4135659661975451,
"recall": 0.6746031746031745,
"f1": 0.4122222222222222
},
"ee": {
"n_samples": 84,
"accuracy": 0.6547619047619048,
"precision": 0.5972649572649572,
"recall": 0.6507936507936507,
"f1": 0.6069243156199677
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9344262295081968,
"precision": 0.7142857142857143,
"recall": 0.9655172413793103,
"f1": 0.7821428571428571
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.2563386788633516,
"js_divergence": 0.06872839042316117
}
}
},
{
"fold": 3,
"best_params": {
"dropout": 0.11061601330629428,
"lr": 0.0003491022701087641,
"weight_decay": 1.2696904821623371e-05,
"backbone_lr_ratio": 0.39277239063286945,
"moe_n_experts": 4,
"moe_top_k": 2,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 32,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 15,
"test_metrics": {
"size": {
"n_samples": 83,
"mse": 0.30582211903190704,
"rmse": 0.5530118615652896,
"mae": 0.297713055668107,
"r2": 0.06157355129088471
},
"delivery": {
"n_samples": 59,
"mse": 0.7117105624010245,
"rmse": 0.8436293987296937,
"mae": 0.5593153651077616,
"r2": 0.30813277436953124
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.36681547619047616,
"recall": 0.5662442396313364,
"f1": 0.3804035250463822
},
"ee": {
"n_samples": 84,
"accuracy": 0.6666666666666666,
"precision": 0.5864396721189867,
"recall": 0.6201608848667672,
"f1": 0.5949003714961162
},
"toxic": {
"n_samples": 60,
"accuracy": 0.9333333333333333,
"precision": 0.7142857142857143,
"recall": 0.9649122807017544,
"f1": 0.7818181818181817
},
"biodist": {
"n_samples": 60,
"kl_divergence": 0.31884722326826903,
"js_divergence": 0.08589330459529161
}
}
},
{
"fold": 4,
"best_params": {
"dropout": 0.41353294032230026,
"lr": 0.00018606616740103076,
"weight_decay": 9.912252717813433e-05,
"backbone_lr_ratio": 0.32820993966799533,
"moe_n_experts": 4,
"moe_top_k": 1,
"moe_expert_hidden_mult": 2,
"llm_lora_r": 16,
"d_model": 256,
"num_heads": 8,
"n_attn_layers": 4,
"fusion_strategy": "attention",
"head_hidden_dim": 128,
"set_transformer_block": "sab"
},
"epoch_mean": 25,
"test_metrics": {
"size": {
"n_samples": 84,
"mse": 0.21199504587721935,
"rmse": 0.4604291974638656,
"mae": 0.33512775670914424,
"r2": -0.5389986893953644
},
"delivery": {
"n_samples": 58,
"mse": 0.8018784776714993,
"rmse": 0.8954766762297605,
"mae": 0.6036076779509413,
"r2": 0.22459674958332065
},
"pdi": {
"n_samples": 84,
"accuracy": 0.6190476190476191,
"precision": 0.4494096489170381,
"recall": 0.7124935995903737,
"f1": 0.47481767481767473
},
"ee": {
"n_samples": 84,
"accuracy": 0.7142857142857143,
"precision": 0.664516711833785,
"recall": 0.7427350427350428,
"f1": 0.6874551971326165
},
"toxic": {
"n_samples": 59,
"accuracy": 0.9661016949152542,
"precision": 0.8,
"recall": 0.9821428571428572,
"f1": 0.8659090909090909
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3693123344696232,
"js_divergence": 0.09443444364888388
}
}
}
],
"summary_stats": {
"size": {
"mse_mean": 0.21249719869892153,
"mse_std": 0.083819203886272,
"rmse_mean": 0.4509291694638765,
"rmse_std": 0.09570832161071456,
"mae_mean": 0.29299021591121593,
"mae_std": 0.04711634555603193,
"r2_mean": -0.33611237791230353,
"r2_std": 0.3452040990881361
},
"delivery": {
"mse_mean": 0.7425693610280117,
"mse_std": 0.15313439940788104,
"rmse_mean": 0.8571056675991812,
"rmse_std": 0.08910238828097625,
"mae_mean": 0.5779881956006352,
"mae_std": 0.040317114677088234,
"r2_mean": 0.24162283519579483,
"r2_std": 0.08163767158765754
},
"pdi": {
"accuracy_mean": 0.634126984126984,
"accuracy_std": 0.08009190991455786,
"precision_mean": 0.36320050940978454,
"precision_std": 0.06018646749200912,
"recall_mean": 0.5239036063335376,
"recall_std": 0.14740021344728976,
"f1_mean": 0.37610884530892047,
"f1_std": 0.0760087699533074
},
"ee": {
"accuracy_mean": 0.6563492063492063,
"accuracy_std": 0.06431020501550125,
"precision_mean": 0.6032694055441358,
"precision_std": 0.07290814030935075,
"recall_mean": 0.6464310864983134,
"recall_std": 0.09580096959112427,
"f1_mean": 0.6100615120789444,
"f1_std": 0.08078656411296503
},
"toxic": {
"accuracy_mean": 0.9511890595408049,
"accuracy_std": 0.016071213375949578,
"precision_mean": 0.7617961570593149,
"precision_std": 0.0689235340767993,
"recall_mean": 0.958294443004062,
"recall_std": 0.05613748087593979,
"f1_mean": 0.8151453855878632,
"f1_std": 0.03136647604932653
},
"biodist": {
"kl_divergence_mean": 0.3412071963471256,
"kl_divergence_std": 0.07500564394419486,
"js_divergence_mean": 0.08684355805526371,
"js_divergence_std": 0.020779098028044035
}
}
}

1
models/abl/llm/20260617_203747/outer_fold_0/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 61, 62, 63, 64, 65, 66, 67, 68, 71, 72, 73, 75, 76, 77, 80, 81, 85, 86, 88, 89, 90, 92, 94, 95, 96, 98, 99, 100, 101, 103, 104, 105, 106, 107, 108, 109, 111, 113, 114, 115, 116, 117, 118, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 133, 134, 135, 136, 138, 139, 141, 144, 145, 146, 147, 148, 149, 150, 151, 153, 154, 155, 156, 157, 158, 159, 160, 163, 164, 167, 168, 169, 170, 171, 172, 175, 176, 178, 179, 180, 182, 183, 184, 185, 188, 189, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 206, 208, 210, 211, 212, 213, 216, 217, 218, 219, 220, 222, 224, 225, 227, 228, 229, 230, 231, 232, 233, 234, 236, 237, 239, 241, 242, 244, 245, 247, 248, 249, 250, 252, 253, 255, 256, 257, 258, 259, 260, 261, 262, 265, 267, 268, 269, 270, 271, 272, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 285, 286, 287, 288, 291, 292, 294, 296, 297, 298, 299, 300, 301, 303, 304, 306, 307, 308, 309, 310, 311, 312, 313, 316, 317, 318, 319, 320, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 333, 335, 336, 337, 338, 339, 340, 342, 343, 344, 345, 346, 347, 348, 349, 351, 352, 354, 355, 356, 357, 358, 359, 360, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372, 374, 375, 377, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391, 393, 395, 396, 397, 398, 399, 400, 401, 403, 404, 405, 406, 407, 408, 409, 411, 412, 413, 414, 415, 417, 418, 419], "outer_test_idx": [7, 16, 20, 29, 51, 59, 60, 69, 70, 74, 78, 79, 82, 83, 84, 87, 91, 93, 97, 102, 110, 112, 119, 120, 121, 132, 137, 140, 142, 143, 152, 161, 162, 165, 166, 173, 174, 177, 181, 186, 187, 194, 205, 207, 209, 214, 215, 221, 223, 226, 235, 238, 240, 243, 246, 251, 254, 263, 264, 266, 273, 284, 289, 290, 293, 295, 302, 305, 314, 315, 331, 334, 341, 350, 353, 361, 373, 376, 378, 392, 394, 402, 410, 416]}

42
models/abl/llm/20260617_203747/outer_fold_0/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 83,
"mse": 0.32257861692567663,
"rmse": 0.567960048705608,
"mae": 0.31035865933062085,
"r2": -0.07107675306042016
},
"delivery": {
"n_samples": 58,
"mse": 0.5663698080597032,
"rmse": 0.7525754500777335,
"mae": 0.5982509058225771,
"r2": 0.28010979698326477
},
"pdi": {
"n_samples": 84,
"accuracy": 0.7619047619047619,
"precision": 0.41369047619047616,
"recall": 0.570647931303669,
"f1": 0.4308211473565804
},
"ee": {
"n_samples": 84,
"accuracy": 0.7261904761904762,
"precision": 0.6878324844368987,
"recall": 0.7209523809523809,
"f1": 0.6887741512487962
},
"toxic": {
"n_samples": 58,
"accuracy": 0.9655172413793104,
"precision": 0.75,
"recall": 0.9821428571428572,
"f1": 0.8242424242424242
},
"biodist": {
"n_samples": 58,
"kl_divergence": 0.3926160201854191,
"js_divergence": 0.08674307332595234
}
}

1
models/abl/llm/20260617_203747/outer_fold_1/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 3, 5, 6, 7, 10, 11, 12, 13, 15, 16, 18, 19, 20, 21, 24, 26, 27, 29, 30, 31, 32, 33, 35, 36, 38, 39, 40, 42, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 65, 66, 67, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 97, 98, 99, 100, 101, 102, 103, 105, 106, 107, 108, 109, 110, 112, 114, 115, 118, 119, 120, 121, 123, 124, 125, 126, 128, 129, 130, 131, 132, 136, 137, 140, 141, 142, 143, 144, 145, 146, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 161, 162, 164, 165, 166, 167, 168, 169, 171, 172, 173, 174, 175, 176, 177, 178, 179, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 194, 195, 196, 197, 198, 199, 201, 202, 204, 205, 206, 207, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238, 240, 242, 243, 244, 246, 247, 248, 250, 251, 252, 254, 255, 256, 258, 259, 260, 261, 262, 263, 264, 266, 268, 269, 270, 271, 273, 274, 275, 276, 277, 278, 279, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 300, 301, 302, 303, 304, 305, 306, 307, 308, 310, 311, 313, 314, 315, 316, 317, 318, 319, 321, 322, 323, 324, 325, 327, 328, 329, 330, 331, 333, 334, 335, 337, 338, 339, 340, 341, 342, 343, 344, 345, 348, 349, 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 361, 362, 364, 365, 366, 367, 368, 369, 373, 374, 375, 376, 378, 379, 380, 381, 382, 384, 387, 388, 389, 391, 392, 393, 394, 396, 397, 398, 399, 402, 403, 404, 405, 406, 408, 409, 410, 411, 412, 413, 414, 416, 417, 419], "outer_test_idx": [1, 2, 4, 8, 9, 14, 17, 22, 23, 25, 28, 34, 37, 41, 43, 45, 63, 64, 68, 85, 96, 104, 111, 113, 116, 117, 122, 127, 133, 134, 135, 138, 139, 147, 160, 163, 170, 180, 191, 192, 193, 200, 203, 208, 224, 227, 230, 239, 241, 245, 249, 253, 257, 265, 267, 272, 280, 294, 298, 299, 309, 312, 320, 326, 332, 336, 346, 347, 360, 363, 370, 371, 372, 377, 383, 385, 386, 390, 395, 400, 401, 407, 415, 418]}

42
models/abl/llm/20260617_203747/outer_fold_1/test_metrics.json
View File

@ -1,42 +0,0 @@
{
"size": {
"n_samples": 84,
"mse": 0.08957809256845296,
"rmse": 0.29929599490880754,
"mae": 0.23231829348064603,
"r2": -0.4840085838962582
},
"delivery": {
"n_samples": 61,
"mse": 1.0507386739143396,
"rmse": 1.0250554491901107,
"mae": 0.6530180171926002,
"r2": 0.20466657047540615
},
"pdi": {
"n_samples": 84,
"accuracy": 0.4880952380952381,
"precision": 0.24191919191919192,
"recall": 0.21825396825396826,
"f1": 0.2222222222222222
},
"ee": {
"n_samples": 84,
"accuracy": 0.5476190476190477,
"precision": 0.49025974025974023,
"recall": 0.4884639170353456,
"f1": 0.4755038065382893
},
"toxic": {
"n_samples": 61,
"accuracy": 0.9508196721311475,
"precision": 0.75,
"recall": 0.9741379310344828,
"f1": 0.8200589970501475
},
"biodist": {
"n_samples": 61,
"kl_divergence": 0.49752328596137474,
"js_divergence": 0.1298567265780671
}
}

1
models/abl/llm/20260617_203747/outer_fold_2/splits.json
View File

@ -1 +0,0 @@
{"outer_train_idx": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 14, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 36, 37, 38, 39, 40, 41, 43, 44, 45, 47, 48, 49, 50, 51, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 68, 69, 70, 72, 74, 76, 78, 79, 80, 82, 83, 84, 85, 86, 87, 88, 91, 93, 94, 95, 96, 97, 100, 101, 102, 103, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 142, 143, 145, 147, 148, 149, 150, 151, 152, 153, 156, 157, 158, 159, 160, 161, 162, 163, 165, 166, 167, 168, 170, 171, 172, 173, 174, 175, 177, 179, 180, 181, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 196, 197, 199, 200, 202, 203, 204, 205, 207, 208, 209, 210, 211, 212, 213, 214, 215, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 232, 233, 235, 236, 237, 238, 239, 240, 241, 242, 243, 245, 246, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 264, 265, 266, 267, 268, 269, 271, 272, 273, 274, 275, 276, 280, 281, 282, 283, 284, 285, 287, 289, 290, 293, 294, 295, 296, 297, 298, 299, 300, 302, 303, 305, 307, 308, 309, 310, 311, 312, 313, 314, 315, 317, 318, 320, 321, 322, 323, 324, 326, 327, 329, 330, 331, 332, 333, 334, 335, 336, 337, 338, 339, 340, 341, 342, 344, 345, 346, 347, 348, 349, 350, 351, 352, 353, 354, 355, 356, 358, 360, 361, 363, 364, 365, 366, 368, 369, 370, 371, 372, 373, 376, 377, 378, 379, 380, 381, 382, 383, 385, 386, 387, 388, 389, 390, 392, 394, 395, 398, 399, 400, 401, 402, 404, 405, 407, 408, 409, 410, 412, 413, 414, 415, 416, 417, 418], "outer_test_idx": [10, 12, 15, 31, 35, 42, 46, 52, 53, 54, 62, 66, 67, 71, 73, 75, 77, 81, 89, 90, 92, 98, 99, 105, 123, 124, 141, 144, 146, 154, 155, 164, 169, 176, 178, 182, 189, 190, 195, 198, 201, 206, 216, 217, 218, 219, 231, 234, 244, 247, 248, 258, 262, 270, 277, 278, 279, 286, 288, 291, 292, 301, 304, 306, 316, 319, 325, 328, 343, 357, 359, 362, 367, 374, 375, 384, 391, 393, 396, 397, 403, 406, 411, 419]}

Some files were not shown because too many files have changed in this diff Show More