def test_fit_generator_with_validation(self):
# declare a batch generator
train, test = train_test_split(self.data, test_size=.2)
train_bg = BatchGenerator(train,
x_structure=('x1', None),
y_structure=('y1', self.y1_enc))
test_bg = BatchGenerator(test,
x_structure=('x1', None),
y_structure=('y1', self.y1_enc))
inp = keras.layers.Input(shape=(1, ))
x = keras.layers.Dense(4, activation='softmax')(inp)
keras_model = keras.models.Model(inp, x)
keras_model.compile('adam', 'sparse_categorical_crossentropy')
keras_model.fit_generator(train_bg, epochs=1, validation_data=test_bg)
def test_mro(self):
mro = BatchGenerator.mro()
i1 = [i for i, x in enumerate(mro) if x == BaseBatchGenerator]
i2 = [i for i, x in enumerate(mro) if x == Sequence]
assert len(i1) == 1
assert len(i2) == 1
# check that BaseBatch generator is resolved before SequenceTF
assert i1 < i2
def test_predict_without_y(self):
keras_bg = BatchGenerator(self.data, x_structure=('x1', None))
inp = keras.layers.Input(shape=(1, ))
x = keras.layers.Dense(4, activation='softmax')(inp)
keras_model = keras.models.Model(inp, x)
keras_model.compile('adam', 'sparse_categorical_crossentropy')
pred = keras_model.predict_generator(keras_bg, verbose=1)
assert type(pred) == np.ndarray
def test_fit_generator_multiple_epochs(self):
# declare a batch generator
keras_bg = BatchGenerator(self.data,
x_structure=('x1', None),
y_structure=('y1', self.y1_enc))
inp = keras.layers.Input(shape=(1, ))
x = keras.layers.Dense(4, activation='softmax')(inp)
keras_model = keras.models.Model(inp, x)
keras_model.compile('adam', 'sparse_categorical_crossentropy')
keras_model.fit_generator(keras_bg, epochs=2)
def test_multiple_inputs(self):
keras_bg = BatchGenerator(self.data,
x_structure=[('x1', None), ('x2', None)],
y_structure=('y1', self.y1_enc))
inp1 = keras.layers.Input(shape=(1, ))
inp2 = keras.layers.Input(shape=(1, ))
x = keras.layers.Concatenate()([inp1, inp2])
x = keras.layers.Dense(4, activation='softmax')(x)
keras_model = keras.models.Model([inp1, inp2], x)
keras_model.compile('adam', 'sparse_categorical_crossentropy')
keras_model.fit_generator(keras_bg, epochs=2)