Merge branch 'main' of https://gitty.informatik.hs-mannheim.de/3016498/ANLP_WS24_CA2

LSTM.py

@@ -6,6 +6,9 @@ from sklearn.metrics import mean_squared_error, r2_score
 from torch.optim.lr_scheduler import ReduceLROnPlateau
 import time
 from tqdm import tqdm
+from Datasets import GloveDataset as HumorDataset
+import Datasets
+import dataset_helper
 
 
 class LSTMNetwork(nn.Module):
@@ -17,7 +20,8 @@ class LSTMNetwork(nn.Module):
 
     def forward(self, x):
         lstm_out, _ = self.lstm(x)
-        return self.fc(self.dropout(lstm_out[:, -1, :]))
+        # print(lstm_out)
+        return self.fc(self.dropout(lstm_out))
 
 
 def compute_metrics(predictions, labels):
@@ -47,8 +51,8 @@ def train_model(model, train_loader, val_loader, test_loader, epochs=10, device=
 
         for batch in tqdm(train_loader, desc=f"Epoch {epoch+1}/{epochs}", ncols=100):
             optimizer.zero_grad()
-            inputs = batch['input_ids'].to(device)
-            labels = batch['labels'].to(device)
+            inputs = batch[0].float().to(device)#batch['input_ids'].to(device)
+            labels = batch[1].float().to(device)#batch['labels'].to(device)
             outputs = model(inputs)
             loss = criterion(outputs.squeeze(), labels)
             loss.backward()
@@ -65,8 +69,8 @@ def train_model(model, train_loader, val_loader, test_loader, epochs=10, device=
 
         with torch.no_grad():
             for batch in val_loader:
-                inputs = batch['input_ids'].to(device)
-                labels = batch['labels'].to(device)
+                inputs = batch[0].float().to(device)#batch['input_ids'].to(device)
+                labels = batch[1].float().to(device)#batch['labels'].to(device)
                 outputs = model(inputs)
                 val_loss += criterion(outputs.squeeze(), labels).item()
                 val_preds.extend(outputs.squeeze().cpu().numpy())
@@ -78,8 +82,8 @@ def train_model(model, train_loader, val_loader, test_loader, epochs=10, device=
 
         with torch.no_grad():
             for batch in test_loader:
-                inputs = batch['input_ids'].to(device)
-                labels = batch['labels'].to(device)
+                inputs = batch[0].float().to(device)#batch['input_ids'].to(device)
+                labels = batch[1].float().to(device)#batch['labels'].to(device)
                 outputs = model(inputs)
                 test_preds.extend(outputs.squeeze().cpu().numpy())
                 test_labels.extend(labels.cpu().numpy())
@@ -101,7 +105,7 @@ def train_model(model, train_loader, val_loader, test_loader, epochs=10, device=
         if test_r2 > best_test_r2:
             best_test_r2 = test_r2
             torch.save(model.state_dict(), "best_lstm_model.pth")
-            print(f"🚀 New best model saved (R2: {test_r2:.4f})")
+            print(f"New best model saved (R2: {test_r2:.4f})")
 
         if avg_val_loss < best_val_loss:
             best_val_loss = avg_val_loss
@@ -109,25 +113,41 @@ def train_model(model, train_loader, val_loader, test_loader, epochs=10, device=
         else:
             counter += 1
             if counter >= patience:
-                print("⛔ Early stopping triggered!")
+                print("Early stopping triggered!")
                 break
 
     return history
 
 
 if __name__ == "__main__":
-    data_path = '/content/drive/MyDrive/Colab Notebooks/ANLP_WS24_CA2/data/embedded_padded'
-    train_dataset = torch.load(f'{data_path}/train.pt')
-    test_dataset = torch.load(f'{data_path}/test.pt')
-    val_dataset = torch.load(f'{data_path}/val.pt')
-
-    input_dim = 100
+    input_dim = 128
     hidden_dim = 1024
     num_layers = 2
     output_dim = 1
     dropout = 0.2
     batch_size = 256
     epochs = 5
+    
+    DATA_PATH = "data/hack.csv"
+    GLOVE_PATH = "data/glove.6b.100d.txt"
+    EMBEDDING_DIM = 100
+    TEST_SIZE = 0.1
+    VAL_SIZE = 0.1
+    params = {"max_len":128}
+    # data_path = 'data/embedded_padded'
+    embedding_matrix, word_index, vocab_size, d_model = dataset_helper.get_embedding_matrix(
+        gloVe_path=GLOVE_PATH, emb_len=EMBEDDING_DIM)
+
+    X, y = dataset_helper.load_preprocess_data(path_data=DATA_PATH, verbose=True)
+
+    # Aufteilen der Daten
+    data_split = dataset_helper.split_data(X, y, test_size=TEST_SIZE, val_size=VAL_SIZE)
+
+    # Dataset und DataLoader
+    train_dataset = Datasets.GloveDataset(data_split['train']['X'], data_split['train']['y'], word_index, max_len=params["max_len"])
+    val_dataset = Datasets.GloveDataset(data_split['val']['X'], data_split['val']['y'], word_index, max_len=params["max_len"])
+    test_dataset = Datasets.GloveDataset(data_split['test']['X'], data_split['test']['y'], word_index, max_len=params["max_len"])
+
 
     train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
     val_loader = DataLoader(val_dataset, batch_size=batch_size, shuffle=False)
@@ -135,5 +155,4 @@ if __name__ == "__main__":
 
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     model = LSTMNetwork(input_dim=input_dim, hidden_dim=hidden_dim, num_layers=num_layers, output_dim=output_dim, dropout=dropout).to(device)
-
     history = train_model(model, train_loader, val_loader, test_loader, epochs=epochs, device=device)