Merge pull request #7 from TailUFPB/neural-network-parameters

Neural network parameters

api/Neural_Network2.py

@@ -8,7 +8,7 @@
 from tensorflow.keras.layers import TextVectorization
 from tensorflow.keras.models import Sequential
 from sklearn.preprocessing import LabelEncoder
-
+import joblib
 def preprocess_text(text):
  text = text.lower()
  text = re.sub('\[.*?\]', '', text)
@@ -20,13 +20,18 @@ def preprocess_text(text):
  text = re.sub('\w*\d\w*', '', text)
  return text
 
-def create_and_train_model(train_texts, train_labels, name, epochs=5):
+def create_and_train_model(train_texts, train_labels, name, epochs=5, batch_size=32):
  label_encoder = LabelEncoder()
  train_labels_encoded = label_encoder.fit_transform(train_labels)
 
  num_classes = len(label_encoder.classes_)
  train_labels_one_hot = tf.keras.utils.to_categorical(train_labels_encoded, num_classes=num_classes)
 
+ # Salva o mapeamento de rótulos em um arquivo
+ label_mapping_filename = f"models/LabelMapping-{name}.joblib"
+ joblib.dump(label_encoder, label_mapping_filename)
+
+
  print(train_texts)
  print(train_labels_one_hot)
 
@@ -76,11 +81,11 @@ def vectorize_text(text, label):
  predictions = layers.Dense(num_classes, activation="softmax", name="predictions")(x)
 
  # Cria e compila o modelo
- model = tf.keras.Model(inputs, predictions)
+ model = tf.keras.Model(inputs, predictions, name)
  model.compile(loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"])
 
  # Treina o modelo
- history = model.fit(train_ds, epochs=epochs)
+ history = model.fit(train_ds, epochs=epochs, batch_size=batch_size)
 
  # Salva o modelo
  model_filename = f"models/Trained-Model-{name}.keras"
@@ -98,6 +103,36 @@ def vectorize_text(text, label):
  except Exception as e:
  return f"Error during model creation/training: {str(e)}"
 
+def predict_with_label_mapping_nn(model_name, test_texts):
+ # Cria uma camada de vetorização de texto
+ vectorize_layer = TextVectorization(
+ max_tokens=20000,
+ output_mode="int",
+ output_sequence_length=500,
+ )
+ # Carrega o modelo treinado
+ model = tf.keras.models.load_model(f"models/Trained-Model-{model_name}.keras")
+
+ # Carrega o mapeamento de rótulos
+ label_mapping_filename = f"models/LabelMapping-{model_name}.joblib"
+ label_encoder = joblib.load(label_mapping_filename)
+
+ # Aplica o pré-processamento aos textos de teste
+ test_texts = [preprocess_text(text) for text in test_texts]
+
+ # Vetoriza os textos de teste usando a mesma camada de vetorização
+ test_ds = vectorize_layer(np.array(test_texts)).numpy()
+
+ # Faz previsões
+ predictions = model.predict(test_ds)
+
+ # Destransforma as previsões usando o mapeamento de rótulos
+ predicted_labels_encoded = np.argmax(predictions, axis=1)
+ predicted_labels = [label_encoder.classes_[label] for label in predicted_labels_encoded]
+
+ return predicted_labels
+
+
 '''
 Com o nome do arquivo podemos fazer por exemplo:
 

api/app.py

@@ -61,7 +61,9 @@ def train_model():
  selected_data = received_data.get('data')
  selected_label = received_data.get('label')
  name = received_data.get('name')
- return create_and_train_model(selected_data,selected_label,name)
+ epochs = received_data.get('epochs')
+ batch_size = received_data.get('batch_size')
+ return create_and_train_model(selected_data, selected_label, name, epochs, batch_size)
 
 if __name__ == '__main__':
  server_thread = threading.Thread(target=run_flask_app)