change probability func

inference.py

@@ -165,7 +165,7 @@ def scaled_sigmoid(x, scale_factor=0.2, linear_property=0.3):
     # Combine sigmoid with linear component
     raw_prob = torch.sigmoid(scaled_x) * (1-linear_property) + linear_property * ((x + 25) / 50)
     # Clip to ensure bounds
-    return torch.clamp(raw_prob, min=0.001, max=0.999)
+    return torch.clamp(raw_prob, min=0.01, max=0.99)
 
 # Apply the scaled sigmoid
 
@@ -175,7 +175,7 @@ def get_model(model_type, device):
     if model_type == "MERT":
         #from model import MusicAudioClassifier
         #model = MusicAudioClassifier(input_dim=768, is_emb=True, mode = 'both', share_parameter = False).to(device)
-        ckpt_file = 'checkpoints/step=007000-val_loss=0.1831-val_acc=0.9278.ckpt'#'mert_finetune_10.pth'
+        ckpt_file = 'checkpoints/step=003432-val_loss=0.0216-val_acc=0.9963.ckpt'#'mert_finetune_10.pth'
         model = MERT_AudioCNN.load_from_checkpoint(ckpt_file).to(device)
         model.eval()
         # model.load_state_dict(torch.load(ckpt_file, map_location=device))
@@ -210,7 +210,7 @@ def inference(audio_path):
 
     # 모델 로드 부분 추가
     model = MusicAudioClassifier.load_from_checkpoint(
-        checkpoint_path = 'checkpoints/EmbeddingModel_MERT_768-epoch=0073-val_loss=0.1058-val_acc=0.9585-val_f1=0.9366-val_precision=0.9936-val_recall=0.8857.ckpt',
+        checkpoint_path = 'checkpoints/EmbeddingModel_MERT_768_2class_weighted-epoch=0014-val_loss=0.0099-val_acc=0.9993-val_f1=0.9978-val_precision=0.9967-val_recall=0.9989.ckpt',
         input_dim=input_dim, 
         #emb_model=backbone_model
     )