fix: sqrt 逆频率权重 + 每 epoch 打印

src/models/train.py

@@ -168,10 +168,10 @@ def train() -> HeatRiskPredictor:
     print(f"模型参数量: {sum(p.numel() for p in model.parameters()):,}")
 
     # -------------------- 损失、优化器、调度器 --------------------
-    # 基于训练集类别分布计算权重
+    # 基于训练集类别分布计算权重 (sqrt inverse freq, 温和加权)
     class_counts = np.bincount(y_train_np[:, 0])
     class_weights_tensor = torch.tensor(
-        1.0 / (class_counts + 1), dtype=torch.float32
+        1.0 / np.sqrt(class_counts + 1), dtype=torch.float32
     ).to(device)
     class_weights_tensor = class_weights_tensor / class_weights_tensor.sum() * 4
     focal_loss = FocalLoss(class_weight=class_weights_tensor)
@@ -256,7 +256,7 @@ def train() -> HeatRiskPredictor:
         scheduler.step(avg_val_loss)
 
         # ---- 打印进度 ----
-        if epoch % 5 == 0 or epoch == 1:
+        if True:  # print every epoch
             lr_now = optimizer.param_groups[0]["lr"]
             print(
                 f"Epoch {epoch:3d}/{MAX_EPOCHS} | "