def fit_neural_network(y_train=None,
X_train=None,
data=None,
layers=None,
n_epochs=100,
n_batch_size=20):
"""Fit a neural network regressor.
Args:
layers (str): str specifying the number of hidden layers and hidden nodes in the
neural network. Example: "100-100-100".
n_epochs (int): Number of epochs used for model fitting.
n_batch_size (int): Batch size used for model fitting.
Returns:
nnet: The fitted neural network.
"""
if data is not None:
y_train = data["y_train"]
X_train = data["X_train"]
build_model = _get_build_model_func(input_dim=X_train.shape[1],
layers=layers)
nnet = KerasClassifier(build_fn=build_model,
batch_size=n_batch_size,
epochs=n_epochs)
nnet._estimator_type = "classifier"
nnet.fit(X_train, y_train, verbose=False)
return nnet
def pima_models(x_train_p, y_train_p):
"""
3 ML models for PIMA (scaled) data
:param x_train_p:
:param Y_train:
:return:
"""
def create_mlp():
mlp = Sequential()
mlp.add(Dense(60, input_dim=len(x_train_p.columns), activation='relu'))
mlp.add(Dropout(0.2))
mlp.add(Dense(30, input_dim=60, activation='relu'))
mlp.add(Dropout(0.2))
mlp.add(Dense(1, activation='sigmoid'))
mlp.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return mlp
mlp = KerasClassifier(build_fn=create_mlp,
epochs=100,
batch_size=64,
verbose=0)
mlp._estimator_type = "classifier"
mlp.fit(x_train_p, y_train_p)
rf = RandomForestClassifier(n_estimators=1000)
rf.fit(x_train_p, y_train_p)
sigmoidRF = CalibratedClassifierCV(
RandomForestClassifier(n_estimators=1000), cv=5, method='sigmoid')
sigmoidRF.fit(x_train_p, y_train_p)
# lr = LogisticRegression(C=1.)
# lr.fit(x_train_p, y_train_p)
svm = SVC(kernel='linear', probability=True)
svm.fit(x_train_p, y_train_p)
# create a dictionary of our models
estimators = [('rf', sigmoidRF), ('mlp', mlp), ('svm', svm)]
# create our voting classifier, inputting our models
ensemble = VotingClassifier(estimators, voting='soft')
ensemble._estimator_type = "classifier"
ensemble.fit(x_train_p, y_train_p)
return {'mlp': mlp, 'rf': sigmoidRF, 'svm': svm, 'ensemble': ensemble}
def build_model3():
model = Sequential()
model.add(Dense(32, activation='relu', input_shape=(784, )))
model.add(Dense(16, activation='relu'))
model.add(Dense(10, activation='softmax'))
model.compile(optimizer='Adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
return model
if __name__ == '__main__':
(x_train, y_train), (x_test, y_test) = load_data()
model1 = KerasClassifier(build_fn=build_model1, epochs=20, batch_size=64)
model1._estimator_type = "classifier"
model2 = KerasClassifier(build_fn=build_model2, epochs=20, batch_size=64)
model2._estimator_type = "classifier"
model3 = KerasClassifier(build_fn=build_model3, epochs=20, batch_size=64)
model3._estimator_type = "classifier"
# if ‘hard’, uses predicted class labels for majority rule voting.
# if ‘soft’, predicts the class label based on the argmax of the
# sums of the predicted probabilities,
# which is recommended for an ensemble of well-calibrated classifiers.
cls = VotingClassifier(estimators=(['model1', model1],
['model2', model2],
['model3', model3]),
voting='hard')
cls.fit(x_train, y_train)
