feat: better choice of hyperparameters

examples/simple_cancer_binary.py

@@ -1,12 +1,9 @@
-from keras.datasets import imdb
-from keras_preprocessing.sequence import pad_sequences
-
 from neuralnetlib.activations import Sigmoid, ReLU
 from neuralnetlib.layers import Activation, Dense
 from neuralnetlib.losses import BinaryCrossentropy
 from neuralnetlib.model import Model
-from neuralnetlib.optimizers import SGD
-from neuralnetlib.metrics import *
+from neuralnetlib.optimizers import Adam
+from neuralnetlib.metrics import accuracy_score, f1_score, recall_score, precision_score
 
 from sklearn.datasets import load_breast_cancer
 from sklearn.model_selection import train_test_split
@@ -19,17 +16,17 @@ def main():
     X, y = data.data, data.target
 
     # 2. Preprocessing
-    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
+    x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
     scaler = StandardScaler()
-    x_train = scaler.fit_transform(X_train)
-    x_test = scaler.transform(X_test)
+    x_train = scaler.fit_transform(x_train)
+    x_test = scaler.transform(x_test)
     y_train = y_train.reshape(-1, 1)
     y_test = y_test.reshape(-1, 1)
 
-    # 3. Model definition
+    # 4. Model definition
     input_neurons = 30  # Cancer dataset has 30 features
-    num_hidden_layers = 2  # Number of hidden layers
-    hidden_neurons = 30  # Number of neurons in each hidden layer
+    num_hidden_layers = 5  # Number of hidden layers
+    hidden_neurons = 100  # Number of neurons in each hidden layer
     output_neurons = 1  # Binary classification
 
     model = Model()
@@ -43,20 +40,20 @@ def main():
     model.add(Dense(hidden_neurons, output_neurons, random_state=42))
     model.add(Activation(Sigmoid()))
 
-    # 4. Model compilation
-    model.compile(loss_function=BinaryCrossentropy(), optimizer=SGD(learning_rate=0.000001))
+    # 5. Model compilation
+    model.compile(loss_function=BinaryCrossentropy(), optimizer=Adam(learning_rate=0.0001))
 
-    # 5. Model training
-    model.train(x_train, y_train, epochs=20, batch_size=128, metrics=[accuracy_score], random_state=42)
+    # 6. Model training
+    model.train(x_train, y_train, epochs=20, batch_size=48, metrics=[accuracy_score], random_state=42)
 
-    # 6. Model evaluation
+    # 7. Model evaluation
     loss = model.evaluate(x_test, y_test)
     print(f'Test loss: {loss}')
 
-    # 7. Model prediction
+    # 8. Model prediction
     y_pred = model.predict(x_test)
 
-    # 8. Printing some metrics
+    # 9. Printing some metrics
     accuracy = accuracy_score(y_pred, y_test)
     precision = precision_score(y_pred, y_test)
     recall = recall_score(y_pred, y_test)
@@ -66,7 +63,7 @@ def main():
     print(f"Precision: {precision}")
     print(f"Recall: {recall}")
     print(f"F1 Score: {f1}")
-    
-    
+
+
 if __name__ == '__main__':
-    main()
+    main()