Add more metrics

data/processed/test_results.json

@@ -1,37 +1,51 @@
 {
   "loss_metrics": {
-    "loss": 2.8661977450052896,
+    "loss": 2.5374555587768555,
-    "loss_size": 0.44916408757368725,
+    "loss_size": 0.1886825958887736,
-    "loss_pdi": 0.5041926403840383,
+    "loss_pdi": 0.45798932512601215,
-    "loss_ee": 0.9021427234013876,
+    "loss_ee": 0.829658567905426,
-    "loss_delivery": 0.5761533578236898,
+    "loss_delivery": 0.4857304096221924,
-    "loss_biodist": 0.4019051690896352,
+    "loss_biodist": 0.5346279243628184,
-    "loss_toxic": 0.03263980595511384,
+    "loss_toxic": 0.04076674363265435,
-    "acc_pdi": 0.7633587786259542,
+    "acc_pdi": 0.7862595419847328,
-    "acc_ee": 0.6641221374045801,
+    "acc_ee": 0.6793893129770993,
-    "acc_toxic": 0.9702970297029703
+    "acc_toxic": 0.9801980198019802
   },
   "detailed_metrics": {
     "size": {
-      "mse": 0.41126506251447736,
+      "mse": 0.1669999969286325,
-      "rmse": 0.6412995107704959,
+      "rmse": 0.4086563310761654,
-      "mae": 0.41415552388095633,
+      "mae": 0.26111859684375066,
-      "r2": -0.9333718010891026
+      "r2": 0.2149270281561566
     },
     "delivery": {
-      "mse": 0.6277965050686476,
+      "mse": 0.5193460523366603,
-      "rmse": 0.7923361061245711,
+      "rmse": 0.7206566813238189,
-      "mae": 0.5387302115022443,
+      "mae": 0.4828052782115008,
-      "r2": 0.24206702565575944
+      "r2": 0.37299826459145
     },
     "pdi": {
-      "accuracy": 0.7633587786259542
+      "accuracy": 0.7862595419847328,
+      "precision": 0.7282763532763532,
+      "recall": 0.6907738095238095,
+      "f1": 0.7041935483870968
     },
     "ee": {
-      "accuracy": 0.6641221374045801
+      "accuracy": 0.6793893129770993,
+      "precision": 0.612247574088644,
+      "recall": 0.6062951496388029,
+      "f1": 0.6069449904342585
     },
     "toxic": {
-      "accuracy": 0.9702970297029703
+      "accuracy": 0.9801980198019802,
+      "precision": 0.5,
+      "recall": 0.4900990099009901,
+      "f1": 0.495
+    },
+    "biodist": {
+      "n_samples": 101,
+      "kl_divergence": 0.2931957937514963,
+      "js_divergence": 0.07706768601895059
     }
   }
 }

lnp_ml/modeling/predict.py

@@ -217,15 +217,31 @@ def test(
     """
     import json
     import numpy as np
+    from scipy.special import rel_entr
     from sklearn.metrics import (
         mean_squared_error,
         mean_absolute_error,
         r2_score,
         accuracy_score,
-        classification_report,
+        precision_score,
+        recall_score,
+        f1_score,
     )
     from lnp_ml.modeling.trainer import validate
     
+    def kl_divergence(p: np.ndarray, q: np.ndarray, eps: float = 1e-10) -> float:
+        """计算 KL 散度 KL(p || q)"""
+        p = np.clip(p, eps, 1.0)
+        q = np.clip(q, eps, 1.0)
+        return float(np.sum(rel_entr(p, q), axis=-1).mean())
+    
+    def js_divergence(p: np.ndarray, q: np.ndarray, eps: float = 1e-10) -> float:
+        """计算 JS 散度"""
+        p = np.clip(p, eps, 1.0)
+        q = np.clip(q, eps, 1.0)
+        m = 0.5 * (p + q)
+        return float(0.5 * (np.sum(rel_entr(p, m), axis=-1) + np.sum(rel_entr(q, m), axis=-1)).mean())
+    
     logger.info(f"Using device: {device}")
     device_obj = torch.device(device)
     
@@ -287,6 +303,9 @@ def test(
             y_pred = np.array(predictions["pdi"])[mask]
             results["detailed_metrics"]["pdi"] = {
                 "accuracy": float(accuracy_score(y_true, y_pred)),
+                "precision": float(precision_score(y_true, y_pred, average="macro", zero_division=0)),
+                "recall": float(recall_score(y_true, y_pred, average="macro", zero_division=0)),
+                "f1": float(f1_score(y_true, y_pred, average="macro", zero_division=0)),
             }
     
     # 分类指标：EE
@@ -299,6 +318,9 @@ def test(
             y_pred = np.array(predictions["ee"])[mask]
             results["detailed_metrics"]["ee"] = {
                 "accuracy": float(accuracy_score(y_true, y_pred)),
+                "precision": float(precision_score(y_true, y_pred, average="macro", zero_division=0)),
+                "recall": float(recall_score(y_true, y_pred, average="macro", zero_division=0)),
+                "f1": float(f1_score(y_true, y_pred, average="macro", zero_division=0)),
             }
     
     # 分类指标：toxic
@@ -309,6 +331,28 @@ def test(
             y_pred = np.array(predictions["toxic"])[mask.values]
             results["detailed_metrics"]["toxic"] = {
                 "accuracy": float(accuracy_score(y_true, y_pred)),
+                "precision": float(precision_score(y_true, y_pred, average="macro", zero_division=0)),
+                "recall": float(recall_score(y_true, y_pred, average="macro", zero_division=0)),
+                "f1": float(f1_score(y_true, y_pred, average="macro", zero_division=0)),
+            }
+    
+    # 分布指标：biodist
+    biodist_cols = [
+        "Biodistribution_lymph_nodes", "Biodistribution_heart", "Biodistribution_liver",
+        "Biodistribution_spleen", "Biodistribution_lung", "Biodistribution_kidney", "Biodistribution_muscle"
+    ]
+    if all(c in test_df.columns for c in biodist_cols):
+        biodist_true = test_df[biodist_cols].values
+        biodist_pred = np.array(predictions["biodist"])
+        # mask: 有效样本是 sum > 0 且无 NaN
+        mask = (biodist_true.sum(axis=1) > 0) & (~np.isnan(biodist_true).any(axis=1))
+        if mask.any():
+            y_true = biodist_true[mask]
+            y_pred = biodist_pred[mask]
+            results["detailed_metrics"]["biodist"] = {
+                "n_samples": int(mask.sum()),
+                "kl_divergence": kl_divergence(y_true, y_pred),
+                "js_divergence": js_divergence(y_true, y_pred),
             }
     
     # 打印结果

lnp_ml/modeling/train_cv.py

@@ -391,13 +391,30 @@ def test(
     
     使用每个 fold 的模型在对应的测试集上评估，然后汇总结果。
     """
+    from scipy.special import rel_entr
     from sklearn.metrics import (
         mean_squared_error,
         mean_absolute_error,
         r2_score,
         accuracy_score,
+        precision_score,
+        recall_score,
+        f1_score,
     )
     
+    def kl_divergence(p: np.ndarray, q: np.ndarray, eps: float = 1e-10) -> float:
+        """计算 KL 散度 KL(p || q)"""
+        p = np.clip(p, eps, 1.0)
+        q = np.clip(q, eps, 1.0)
+        return float(np.sum(rel_entr(p, q), axis=-1).mean())
+    
+    def js_divergence(p: np.ndarray, q: np.ndarray, eps: float = 1e-10) -> float:
+        """计算 JS 散度"""
+        p = np.clip(p, eps, 1.0)
+        q = np.clip(q, eps, 1.0)
+        m = 0.5 * (p + q)
+        return float(0.5 * (np.sum(rel_entr(p, m), axis=-1) + np.sum(rel_entr(q, m), axis=-1)).mean())
+    
     logger.info(f"Using device: {device}")
     device = torch.device(device)
     
@@ -413,10 +430,10 @@ def test(
     fold_results = []
     # 用于汇总所有 fold 的预测
     all_preds = {
-        "size": [], "delivery": [], "pdi": [], "ee": [], "toxic": []
+        "size": [], "delivery": [], "pdi": [], "ee": [], "toxic": [], "biodist": []
     }
     all_targets = {
-        "size": [], "delivery": [], "pdi": [], "ee": [], "toxic": []
+        "size": [], "delivery": [], "pdi": [], "ee": [], "toxic": [], "biodist": []
     }
     
     for fold_dir in tqdm(fold_dirs, desc="Evaluating folds"):
@@ -522,6 +539,14 @@ def test(
                     toxic_targets = targets["toxic"][mask].cpu().numpy().astype(int)
                     fold_preds["toxic"].extend(toxic_preds.tolist())
                     fold_targets["toxic"].extend(toxic_targets.tolist())
+                
+                # Biodist (distribution)
+                if "biodist" in masks and masks["biodist"].any():
+                    mask = masks["biodist"]
+                    biodist_preds = outputs["biodist"][mask].cpu().numpy()
+                    biodist_targets = targets["biodist"][mask].cpu().numpy()
+                    fold_preds["biodist"].extend(biodist_preds.tolist())
+                    fold_targets["biodist"].extend(biodist_targets.tolist())
         
         # 计算当前 fold 的指标
         fold_metrics = {"fold_idx": fold_idx, "n_samples": len(test_df)}
@@ -546,8 +571,21 @@ def test(
                 fold_metrics[task] = {
                     "n": len(p),
                     "accuracy": float(accuracy_score(t, p)),
+                    "precision": float(precision_score(t, p, average="macro", zero_division=0)),
+                    "recall": float(recall_score(t, p, average="macro", zero_division=0)),
+                    "f1": float(f1_score(t, p, average="macro", zero_division=0)),
                 }
         
+        # 分布任务指标
+        if fold_preds["biodist"]:
+            p = np.array(fold_preds["biodist"])
+            t = np.array(fold_targets["biodist"])
+            fold_metrics["biodist"] = {
+                "n": len(p),
+                "kl_divergence": kl_divergence(t, p),
+                "js_divergence": js_divergence(t, p),
+            }
+        
         fold_results.append(fold_metrics)
         
         # 汇总到全局
@@ -564,6 +602,10 @@ def test(
         for task in ["pdi", "ee", "toxic"]:
             if task in fold_metrics and isinstance(fold_metrics[task], dict):
                 log_parts.append(f"{task}_acc={fold_metrics[task]['accuracy']:.4f}")
+                log_parts.append(f"{task}_f1={fold_metrics[task]['f1']:.4f}")
+        if "biodist" in fold_metrics and isinstance(fold_metrics["biodist"], dict):
+            log_parts.append(f"biodist_KL={fold_metrics['biodist']['kl_divergence']:.4f}")
+            log_parts.append(f"biodist_JS={fold_metrics['biodist']['js_divergence']:.4f}")
         logger.info(", ".join(log_parts))
     
     # 计算跨 fold 汇总统计
@@ -581,12 +623,26 @@ def test(
     
     for task in ["pdi", "ee", "toxic"]:
         accs = [r[task]["accuracy"] for r in fold_results if task in r and isinstance(r[task], dict)]
+        f1s = [r[task]["f1"] for r in fold_results if task in r and isinstance(r[task], dict)]
         if accs:
             summary_stats[task] = {
                 "accuracy_mean": float(np.mean(accs)),
                 "accuracy_std": float(np.std(accs)),
+                "f1_mean": float(np.mean(f1s)),
+                "f1_std": float(np.std(f1s)),
             }
     
+    # 分布任务汇总
+    kls = [r["biodist"]["kl_divergence"] for r in fold_results if "biodist" in r and isinstance(r["biodist"], dict)]
+    jss = [r["biodist"]["js_divergence"] for r in fold_results if "biodist" in r and isinstance(r["biodist"], dict)]
+    if kls:
+        summary_stats["biodist"] = {
+            "kl_mean": float(np.mean(kls)),
+            "kl_std": float(np.std(kls)),
+            "js_mean": float(np.mean(jss)),
+            "js_std": float(np.std(jss)),
+        }
+    
     # 计算整体 pooled 指标
     overall = {}
     for task in ["size", "delivery"]:
@@ -608,8 +664,21 @@ def test(
             overall[task] = {
                 "n_samples": len(p),
                 "accuracy": float(accuracy_score(t, p)),
+                "precision": float(precision_score(t, p, average="macro", zero_division=0)),
+                "recall": float(recall_score(t, p, average="macro", zero_division=0)),
+                "f1": float(f1_score(t, p, average="macro", zero_division=0)),
             }
     
+    # 分布任务
+    if all_preds["biodist"]:
+        p = np.array(all_preds["biodist"])
+        t = np.array(all_targets["biodist"])
+        overall["biodist"] = {
+            "n_samples": len(p),
+            "kl_divergence": kl_divergence(t, p),
+            "js_divergence": js_divergence(t, p),
+        }
+    
     # 打印汇总结果
     logger.info("\n" + "=" * 60)
     logger.info("CV TEST EVALUATION RESULTS")
@@ -619,15 +688,19 @@ def test(
     for task, stats in summary_stats.items():
         if "rmse_mean" in stats:
             logger.info(f"  {task}: RMSE={stats['rmse_mean']:.4f}±{stats['rmse_std']:.4f}, R²={stats['r2_mean']:.4f}±{stats['r2_std']:.4f}")
-        else:
+        elif "accuracy_mean" in stats:
-            logger.info(f"  {task}: Accuracy={stats['accuracy_mean']:.4f}±{stats['accuracy_std']:.4f}")
+            logger.info(f"  {task}: Accuracy={stats['accuracy_mean']:.4f}±{stats['accuracy_std']:.4f}, F1={stats['f1_mean']:.4f}±{stats['f1_std']:.4f}")
+        elif "kl_mean" in stats:
+            logger.info(f"  {task}: KL={stats['kl_mean']:.4f}±{stats['kl_std']:.4f}, JS={stats['js_mean']:.4f}±{stats['js_std']:.4f}")
     
     logger.info(f"\n[Overall (all samples pooled)]")
     for task, metrics in overall.items():
         if "rmse" in metrics:
             logger.info(f"  {task} (n={metrics['n_samples']}): RMSE={metrics['rmse']:.4f}, MAE={metrics['mae']:.4f}, R²={metrics['r2']:.4f}")
-        else:
+        elif "accuracy" in metrics:
-            logger.info(f"  {task} (n={metrics['n_samples']}): Accuracy={metrics['accuracy']:.4f}")
+            logger.info(f"  {task} (n={metrics['n_samples']}): Accuracy={metrics['accuracy']:.4f}, Precision={metrics['precision']:.4f}, Recall={metrics['recall']:.4f}, F1={metrics['f1']:.4f}")
+        elif "kl_divergence" in metrics:
+            logger.info(f"  {task} (n={metrics['n_samples']}): KL={metrics['kl_divergence']:.4f}, JS={metrics['js_divergence']:.4f}")
     
     # 保存结果
     results = {

models/finetune_cv/test_results.json

@@ -17,15 +17,29 @@
       },
       "pdi": {
         "n": 95,
-        "accuracy": 0.6105263157894737
+        "accuracy": 0.6105263157894737,
+        "precision": 0.20350877192982456,
+        "recall": 0.3333333333333333,
+        "f1": 0.25272331154684097
       },
       "ee": {
         "n": 95,
-        "accuracy": 0.6631578947368421
+        "accuracy": 0.6631578947368421,
+        "precision": 0.22580645161290322,
+        "recall": 0.328125,
+        "f1": 0.267515923566879
       },
       "toxic": {
         "n": 66,
-        "accuracy": 0.8939393939393939
+        "accuracy": 0.8939393939393939,
+        "precision": 0.44696969696969696,
+        "recall": 0.5,
+        "f1": 0.472
+      },
+      "biodist": {
+        "n": 66,
+        "kl_divergence": 0.8735123101690443,
+        "js_divergence": 0.2203766048579219
       }
     },
     {
@@ -45,15 +59,29 @@
       },
       "pdi": {
         "n": 195,
-        "accuracy": 0.7076923076923077
+        "accuracy": 0.7076923076923077,
+        "precision": 0.35384615384615387,
+        "recall": 0.5,
+        "f1": 0.4144144144144144
       },
       "ee": {
         "n": 195,
-        "accuracy": 0.4205128205128205
+        "accuracy": 0.4205128205128205,
+        "precision": 0.14017094017094017,
+        "recall": 0.3333333333333333,
+        "f1": 0.19735258724428398
       },
       "toxic": {
         "n": 123,
-        "accuracy": 1.0
+        "accuracy": 1.0,
+        "precision": 1.0,
+        "recall": 1.0,
+        "f1": 1.0
+      },
+      "biodist": {
+        "n": 123,
+        "kl_divergence": 1.218490162626268,
+        "js_divergence": 0.3069412988090679
       }
     },
     {
@@ -73,15 +101,29 @@
       },
       "pdi": {
         "n": 51,
-        "accuracy": 0.8823529411764706
+        "accuracy": 0.8823529411764706,
+        "precision": 0.29411764705882354,
+        "recall": 0.3333333333333333,
+        "f1": 0.3125
       },
       "ee": {
         "n": 51,
-        "accuracy": 0.8431372549019608
+        "accuracy": 0.8431372549019608,
+        "precision": 0.28104575163398693,
+        "recall": 0.3333333333333333,
+        "f1": 0.3049645390070922
       },
       "toxic": {
         "n": 47,
-        "accuracy": 0.851063829787234
+        "accuracy": 0.851063829787234,
+        "precision": 0.425531914893617,
+        "recall": 0.5,
+        "f1": 0.4597701149425288
+      },
+      "biodist": {
+        "n": 45,
+        "kl_divergence": 1.0902737810034395,
+        "js_divergence": 0.27372590260311874
       }
     },
     {
@@ -101,15 +143,29 @@
       },
       "pdi": {
         "n": 66,
-        "accuracy": 0.8484848484848485
+        "accuracy": 0.8484848484848485,
+        "precision": 0.42424242424242425,
+        "recall": 0.5,
+        "f1": 0.4590163934426229
       },
       "ee": {
         "n": 66,
-        "accuracy": 0.18181818181818182
+        "accuracy": 0.18181818181818182,
+        "precision": 0.203921568627451,
+        "recall": 0.2707070707070707,
+        "f1": 0.12297410192147035
       },
       "toxic": {
         "n": 62,
-        "accuracy": 1.0
+        "accuracy": 1.0,
+        "precision": 1.0,
+        "recall": 1.0,
+        "f1": 1.0
+      },
+      "biodist": {
+        "n": 62,
+        "kl_divergence": 0.9434472801013084,
+        "js_divergence": 0.20391570021898991
       }
     },
     {
@@ -129,15 +185,29 @@
       },
       "pdi": {
         "n": 27,
-        "accuracy": 0.8888888888888888
+        "accuracy": 0.8888888888888888,
+        "precision": 0.4444444444444444,
+        "recall": 0.5,
+        "f1": 0.47058823529411764
       },
       "ee": {
         "n": 27,
-        "accuracy": 0.5925925925925926
+        "accuracy": 0.5925925925925926,
+        "precision": 0.19753086419753085,
+        "recall": 0.3333333333333333,
+        "f1": 0.24806201550387597
       },
       "toxic": {
         "n": 15,
-        "accuracy": 1.0
+        "accuracy": 1.0,
+        "precision": 1.0,
+        "recall": 1.0,
+        "f1": 1.0
+      },
+      "biodist": {
+        "n": 15,
+        "kl_divergence": 0.7615404990024607,
+        "js_divergence": 0.20916908426734354
       }
     }
   ],
@@ -156,15 +226,27 @@
     },
     "pdi": {
       "accuracy_mean": 0.7875890604063979,
-      "accuracy_std": 0.11016791908756088
+      "accuracy_std": 0.11016791908756088,
+      "f1_mean": 0.3818484709395992,
+      "f1_std": 0.08529090446864619
     },
     "ee": {
       "accuracy_mean": 0.5402437489124795,
-      "accuracy_std": 0.22467627690136344
+      "accuracy_std": 0.22467627690136344,
+      "f1_mean": 0.2281738334487203,
+      "f1_std": 0.063019179670124
     },
     "toxic": {
       "accuracy_mean": 0.9490006447453256,
-      "accuracy_std": 0.06391582554207781
+      "accuracy_std": 0.06391582554207781,
+      "f1_mean": 0.7863540229885058,
+      "f1_std": 0.26169039387919035
+    },
+    "biodist": {
+      "kl_mean": 0.9774528065805042,
+      "kl_std": 0.1608761675751487,
+      "js_mean": 0.24282571815128842,
+      "js_std": 0.04053726747029075
     }
   },
   "overall": {
@@ -184,15 +266,29 @@
     },
     "pdi": {
       "n_samples": 434,
-      "accuracy": 0.7396313364055299
+      "accuracy": 0.7396313364055299,
+      "precision": 0.18490783410138248,
+      "recall": 0.25,
+      "f1": 0.21258278145695364
     },
     "ee": {
       "n_samples": 434,
-      "accuracy": 0.4976958525345622
+      "accuracy": 0.4976958525345622,
+      "precision": 0.21063404810247777,
+      "recall": 0.2839160839160839,
+      "f1": 0.23684873775319115
     },
     "toxic": {
       "n_samples": 313,
-      "accuracy": 0.9552715654952076
+      "accuracy": 0.9552715654952076,
+      "precision": 0.4776357827476038,
+      "recall": 0.5,
+      "f1": 0.48856209150326796
+    },
+    "biodist": {
+      "n_samples": 311,
+      "kl_divergence": 1.0498561462079121,
+      "js_divergence": 0.25851000311532496
     }
   }
 }