opt = keras.optimizers.Adam(lr=0.001)
if TRAIN:
model.compile(loss='mse',
optimizer=opt,
metrics=[auxilary.coefficientofdetermination])
model.fit(X_train,
y_train,
validation_data=(X_val, y_val),
epochs=2000,
batch_size=32,
callbacks=[es, mc])
bestModel = load_model('best_model.h5',
custom_objects={
'coefficientofdetermination':
auxilary.coefficientofdetermination
})
del X_test['id']
X_test = scaler.fit_transform(X_test)
X_test = np.nan_to_num(X_test, nan=0)
X_test = featureSelection.transform(X_test)
y_predictions = bestModel.predict(X_test)
y_predictions = np.reshape(y_predictions, y_predictions.shape[0])
auxilary.createSubmissionFiles(y_predictions)
model.predict(X_test)
parameters = {
'objective': ['binary:logistic'],
'max_depth': [10],
'min_child_weight': [11],
'n_estimators': [400],
'seed': [1111],
'learning_rate': [0.05],
'max_delta_step': [3],
'num_class': [4]
}
clf = GridSearchCV(estimator=xgb_model,
param_grid=parameters,
n_jobs=5,
cv=10,
scoring=scoreFunction,
verbose=2)
clf.fit(X_train, y_train)
print("Best score of best on validation set: ", clf.best_score_) #0.6670
print("Best Parameters: ", clf.best_params_) #rbf, 10
X_test = scaler.transform(X_test)
if FEATURE_SELECTION:
X_test = featureSelection.transform(X_test)
y_pred_test = clf.predict(X_test)
print('Number of 3:', np.count_nonzero(y_pred_test == 3))
auxilary.createSubmissionFiles(y_pred_test)