def test_saveload_model(self):
# make and train a very small model
inputs = Input(shape=(2, ))
dense1 = Dense(10, activation='relu', name='FC_1')(inputs)
dense2 = Dense(5, activation='relu', name='FC_2')(dense1)
output1 = Dense(1, name='output1')(dense2)
output2 = Dense(1, name='output2')(dense2)
output3 = Dense(2, activation='softmax', name='output3')(dense2)
model = Model(inputs, [output1, output2, output3])
# v1 = np.random.normal(0, 2, 100)
# v2 = np.random.normal(0.3, 0.5, 100)
# v3 = np.random.normal(-0.3, 0.5, 100)
# X = np.array([v1, v2]).T
# y = [v3, v3]
model.compile(optimizer='adam',
loss='mean_squared_error',
metrics=['mae'])
# history = model.fit(X, y, epochs=1, batch_size=20, shuffle=True)
try:
fileio.safe_model(model, {}, [], [], [], {}, 'test.h5')
model_new, _, _, _, _, _, _ = fileio.load_model('test.h5')
finally:
os.remove('test.h5')
assert model.to_json() == model_new.to_json()
def test_saveload_history(self):
# the save function NEEDS a model to work
model = Sequential([
Dense(32, input_shape=(5, )),
Activation('relu'),
Dense(10),
Activation('softmax'),
])
data = {
'M1final_mae': [0.3, 0.2],
'val_M1final_mae': [0.31, 0.21],
'M1final_loss': [1.5, 1.3],
'val_M1final_loss': [1.6, 1.4],
'training_run': [1, 1]
}
history = pd.DataFrame(data=data)
history.index.name = 'epoch'
try:
fileio.safe_model(model, {}, [], [], [], {},
'test.h5',
history=history)
_, _, _, _, _, _, history_new = fileio.load_model('test.h5')
finally:
os.remove('test.h5')
np.testing.assert_array_equal(history.columns, history_new.columns)
np.testing.assert_array_equal(history.values, history_new.values)
assert history_new.index.name == 'epoch'
def save_model(self, filename, include_history=False):
"""
Save a trained model to hdf5 file for later use
"""
history = self.history if include_history else None
fileio.safe_model(self.model,
self.processors,
self.features,
self.regressors,
self.classifiers,
self.setup,
filename,
history=history)
def save_model(self, filename):
"""
Save a trained model to hdf5 file for later use
"""
fileio.safe_model(self.model,
self.processors,
self.features,
self.regressors,
self.classifiers,
self.setup,
filename,
history=None,
method='pickle')
def save_model(self, filename):
"""
Save a trained model to a pickled file for later use.
Gradient boosted models can for now not be saved to hdf5 format.
"""
fileio.safe_model(self.model,
self.processors,
self.features,
self.regressors,
self.classifiers,
self.setup,
filename,
history=None,
method='pickle')