Fallback

lnp_ml/modeling/benchmark.py

@@ -1,7 +1,6 @@
 """Benchmark 脚本：在 baseline 论文公开的 CV 划分上评估模型（仅 delivery 任务）"""
 
 import json
-import math
 from pathlib import Path
 from typing import Dict, List, Optional
 
@@ -10,7 +9,6 @@ import pandas as pd
 import torch
 import torch.nn as nn
 from torch.utils.data import DataLoader
-from torch.optim.lr_scheduler import LambdaLR, CosineAnnealingLR, SequentialLR
 from loguru import logger
 from tqdm import tqdm
 from sklearn.metrics import mean_squared_error, r2_score
@@ -160,7 +158,6 @@ def train_fold(
     weight_decay: float = 1e-5,
     epochs: int = 50,
     patience: int = 10,
-    warmup_epochs: int = 3,
     config: Optional[Dict] = None,
 ) -> Dict:
     """训练单个 fold"""
@@ -169,19 +166,9 @@ def train_fold(
     logger.info(f"{'='*60}")
     
     optimizer = torch.optim.AdamW(model.parameters(), lr=lr, weight_decay=weight_decay)
-    
-    warmup_scheduler = LambdaLR(
-        optimizer, lr_lambda=lambda epoch: (epoch + 1) / warmup_epochs
+    scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
+        optimizer, mode="min", factor=0.5, patience=5
     )
-    cosine_scheduler = CosineAnnealingLR(
-        optimizer, T_max=epochs - warmup_epochs
-    )
-    scheduler = SequentialLR(
-        optimizer,
-        schedulers=[warmup_scheduler, cosine_scheduler],
-        milestones=[warmup_epochs],
-    )
-    
     early_stopping = EarlyStopping(patience=patience)
     
     best_val_loss = float("inf")
@@ -211,7 +198,7 @@ def train_fold(
             "lr": current_lr,
         })
         
-        scheduler.step()
+        scheduler.step(val_metrics["loss"])
         
         if val_metrics["loss"] < best_val_loss:
             best_val_loss = val_metrics["loss"]