add prefix to model path

app.py

@@ -60,7 +60,7 @@ def train():
 def training():
     global train_from
     prepare_training_data(root_folder,database_file,train_from)
-    train_predictor(root_folder,train_from)
+    train_predictor(root_folder,database_file,train_from)
     predict_score(root_folder,database_file,train_from)
     validate_prediction(root_folder,database_file,train_from)
     return redirect('/')

predict_score.py

@@ -10,11 +10,12 @@
 
 
 def predict_score(root_folder, database_file, train_from, clip_model="ViT-L/14"):
+    prefix = database_file.split(".")[0]
     path = pathlib.Path(root_folder)
     database_path = path / database_file
     database = pd.read_csv(database_path)
     model = MLP(768)  # CLIP embedding dim is 768 for CLIP ViT L 14
-    model_name = f"linear_predictor_{clip_model.replace('/', '').lower()}_{train_from}_mse.pth"
+    model_name = f"{prefix}_linear_predictor_{clip_model.replace('/', '').lower()}_{train_from}_mse.pth"
     s = torch.load(path / model_name)   # load the model you trained previously or the model available in this repo
     model.load_state_dict(s)
     model.to("cuda")

train_predictor.py

@@ -52,12 +52,13 @@ def configure_optimizers(self):
         return optimizer
 
 
-def train_predictor(root_folder, train_from, clip_model="ViT-L/14", val_percentage=0.05, epochs=50, batch_size=256):
+def train_predictor(root_folder, database_file, train_from, clip_model="ViT-L/14", val_percentage=0.05, epochs=50, batch_size=256):
 
+    prefix = database_file.split(".")[0]
     out_path = pathlib.Path(root_folder)
-    x_out = f"x_{clip_model.replace('/', '').lower()}_ebeddings.npy"
-    y_out = f"y_{train_from}.npy"
-    save_name = f"linear_predictor_{clip_model.replace('/', '').lower()}_{train_from}_mse.pth"
+    x_out = f"{prefix}_x_{clip_model.replace('/', '').lower()}_ebeddings.npy"
+    y_out = f"{prefix}_y_{train_from}.npy"
+    save_name = f"{prefix}_linear_predictor_{clip_model.replace('/', '').lower()}_{train_from}_mse.pth"
 
     x = np.load(out_path / x_out)
     y = np.load(out_path / y_out)