fix: normalize hyper params

config/pipeline.development.yaml

@@ -29,7 +29,7 @@ llm:
 # Data handling configuration
 data:
   split_evaluation: false
-  max_dataset_size: null
+  max_dataset_size: 100_000
   balance_by_sex: true
 
 # Enhanced logging for development

research/models/bigru_model.py

@@ -13,7 +13,7 @@ from research.neural_network_model import NeuralNetworkModel
 class BiGRUModel(NeuralNetworkModel):
     """Bidirectional GRU model for name classification"""
 
-    def build_model_with_vocab(self, vocab_size: int, max_len: int = 6, **kwargs) -> Any:
+    def build_model_with_vocab(self, vocab_size: int, **kwargs) -> Any:
         params = kwargs
         model = Sequential(
             [
@@ -22,7 +22,6 @@ class BiGRUModel(NeuralNetworkModel):
                 Embedding(
                     input_dim=vocab_size,
                     output_dim=params.get("embedding_dim", 64),
-                    input_length=max_len,
                     mask_zero=True,
                 ),
                 # First recurrent block returns full sequences to allow stacking.

research/models/cnn_model.py

@@ -21,7 +21,7 @@ from research.neural_network_model import NeuralNetworkModel
 class CNNModel(NeuralNetworkModel):
     """1D Convolutional Neural Network for character patterns"""
 
-    def build_model_with_vocab(self, vocab_size: int, max_len: int = 20, **kwargs) -> Any:
+    def build_model_with_vocab(self, vocab_size: int, **kwargs) -> Any:
         """Build CNN model with known vocabulary size"""
 
         params = kwargs

research/models/lstm_model.py

@@ -13,7 +13,7 @@ from research.neural_network_model import NeuralNetworkModel
 class LSTMModel(NeuralNetworkModel):
     """LSTM model for sequence learning"""
 
-    def build_model_with_vocab(self, vocab_size: int, max_len: int = 6, **kwargs) -> Any:
+    def build_model_with_vocab(self, vocab_size: int, **kwargs) -> Any:
         params = kwargs
         model = Sequential(
             [
@@ -21,7 +21,6 @@ class LSTMModel(NeuralNetworkModel):
                 Embedding(
                     input_dim=vocab_size,
                     output_dim=params.get("embedding_dim", 64),
-                    input_length=max_len,
                     mask_zero=True,
                 ),
                 # Stacked bidirectional LSTMs: first returns sequences to feed the next.

research/models/naive_bayes_model.py

@@ -17,7 +17,7 @@ class NaiveBayesModel(TraditionalModel):
         # includes unigrams for coverage and higher n for suffix/prefix cues.
         vectorizer = CountVectorizer(
             analyzer="char",
-            ngram_range=params.get("ngram_range", (1, 4)),
+            ngram_range=params.get("ngram_range", (2, 5)),
             max_features=params.get("max_features", 8000),
         )
 

research/models/transformer_model.py

@@ -22,21 +22,21 @@ from research.neural_network_model import NeuralNetworkModel
 class TransformerModel(NeuralNetworkModel):
     """Transformer-based model"""
 
-    def build_model_with_vocab(self, vocab_size: int, max_len: int = 6, **kwargs) -> Any:
+    def build_model_with_vocab(self, vocab_size: int, **kwargs) -> Any:
         params = kwargs
 
         # Build Transformer model
-        inputs = Input(shape=(max_len,))
+        inputs = Input(shape=(params.get("max_len", 8),))
         x = Embedding(
             input_dim=vocab_size,
             output_dim=params.get("embedding_dim", 64),
-            input_length=max_len,
+            input_length=params.get("max_len", 8),
             mask_zero=True,
         )(inputs)
 
         # Add positional encoding
-        positions = tf.range(start=0, limit=max_len, delta=1)
+        positions = tf.range(start=0, limit=params.get("max_len", 8), delta=1)
-        pos_embedding = Embedding(input_dim=max_len, output_dim=params.get("embedding_dim", 64))(
+        pos_embedding = Embedding(input_dim=params.get("max_len", 8), output_dim=params.get("embedding_dim", 64))(
             positions
         )
         x = x + pos_embedding

research/neural_network_model.py

@@ -52,22 +52,21 @@ class NeuralNetworkModel(BaseModel):
         logging.info(f"Vocabulary size: {vocab_size}")
 
         # Get additional model parameters
-        max_len = self.config.model_params.get("max_len", 6)
+        self.model = self.build_model_with_vocab(vocab_size=vocab_size, **self.config.model_params)
-
-        self.model = self.build_model_with_vocab(
-            vocab_size=vocab_size, max_len=max_len, **self.config.model_params
-        )
 
         # Train the neural network
         logging.info(
             f"Fitting model with {X_prepared.shape[0]} samples and {X_prepared.shape[1]} features"
         )
+        logging.info(X_prepared[0])
+        logging.info(f"Model parameters: {self.config.model_params}")
+
         history = self.model.fit(
             X_prepared,
             y_encoded,
             epochs=self.config.model_params.get("epochs", 10),
             batch_size=self.config.model_params.get("batch_size", 64),
-            validation_split=0.1,
+            validation_split=self.config.model_params.get("validation_split", 0.1),
             verbose=2,
         )
 

research/traditional_model.py

@@ -59,9 +59,19 @@ class TraditionalModel(BaseModel):
                 f"Fitting model with {X_prepared.shape[0]} samples and {X_prepared.shape[1]} features"
             )
 
-        self.model.fit(X_prepared, y_encoded)
+        logging.info(X_prepared[0])
+        logging.info(f"Model parameters: {self.config.model_params}")
+
+        history = self.model.fit(X_prepared, y_encoded)
         self.is_fitted = True
 
+        self.training_history = {
+            "accuracy": history.history["accuracy"],
+            "loss": history.history["loss"],
+            "val_accuracy": history.history.get("val_accuracy", []),
+            "val_loss": history.history.get("val_loss", []),
+        }
+
         return self
 
     def cross_validate(self, X: pd.DataFrame, y: pd.Series, cv_folds: int = 5) -> Dict[str, float]: