def training_curve(X, y, model):
def mlp_databalancing(_X, _y, sampling_strategy):
if sampling_strategy == SamplingStrategy.UNDERSAMPLING:
_X, _y = RandomUnderSampler(random_state=SEED).fit_resample(_X, _y)
elif sampling_strategy == SamplingStrategy.OVERSAMPLING:
_X, _y = SMOTE(random_state=SEED, n_jobs=-1).fit_resample(_X, _y)
return _X, _y
SPLIT_AND_VALIDATE = True # If False, train with all data
SAMPLING_STRATEGY = SamplingStrategy.OVERSAMPLING
with timing():
# Custom code for databalancing for MLP, since imblearn pipeline don't return keras-history
if SPLIT_AND_VALIDATE:
X_train, X_val, y_train, y_val = train_test_split(
X, y, test_size=0.1, random_state=SEED)
if SAMPLING_STRATEGY != SamplingStrategy.NONE:
X_train, y_train = mlp_databalancing(
X_train, y_train, SAMPLING_STRATEGY)
history = model.fit(X_train, y_train, epochs=num_epochs,
batch_size=batch_size, validation_data=(X_val, y_val))
else:
X, y = mlp_databalancing(X, y, SAMPLING_STRATEGY)
history = model.fit(X, y, epochs=num_epochs,
batch_size=batch_size, verbose=0)
plot_learning_curve_keras(history)
def evaluate(X, y, model):
with timing():
scores, averages = evaluate_model(X, y, model)
print(scores)
print(f'Averages: {averages}')
confusion_matrix(X, y, model, XGBOOST_VISUALIZATION_PATH)
def evaluate(X, y):
with timing():
model_for_evaluation = KerasClassifier(
build_fn=create_model, epochs=num_epochs, batch_size=batch_size, verbose=0)
model = create_pipeline(
model_for_evaluation, sampling_strategy=SamplingStrategy.OVERSAMPLING, y=y)
scores, averages = evaluate_model(X, y, model, gpu_mode=True)
print('\n\n', scores)
print(f'Averages: {averages}')
confusion_matrix(X, y, model, MLP_VISUALIZATION_PATH)
def training_curve(X, y, model):
with timing():
plot_learning_curve(X, y, model, XGBOOST_VISUALIZATION_PATH)