def test_repeat_block_sequence(self):
Dense('256', input_shape='784').b(self.b1).run()
Dropout('0.5').run()
RepeatBegin().b(self.b2).run()
Dense('128', activation='relu').b(self.b1).run()
Dropout('0.2').run()
Dense('128', activation='sigmoid').run()
Dropout('0.8').run()
RepeatEnd(self.b2, '2').run()
Dense('10', activation='softmax').run()
SummaryLayer(self.b1).run()
def build_model():
m = MODEL_BRANCH
# build model
InputLayer('784').b(m).run()
Dense('512', activation='relu').run()
# Dense('512', activation='relu', input_shape='784').b(m).run()
Dropout('0.2').run()
Dense('512', activation='relu').run()
Dropout('0.2').run()
Dense('10', activation='softmax').run()
# compile model
Compile('categorical_crossentropy', 'rmsprop', ['accuracy']).run()
# show network struct
SummaryLayer(m).run()
def setUp(self) -> None:
n1 = BRANCH_1
n2 = BRANCH_2
# first branch
InputLayer('784').b(n1).run()
Dense('256').run()
Dropout('0.5').run()
Dense('64').run()
Dropout('0.2').run()
# second branch
InputLayer('256').b(n2).run()
Dense('64').run()
Dropout('0.5').run()
def test_repeat_block_network_branch(self):
# TODO 暂不兼容复用分支快包含多分支类型
InputLayer(input_shape='784').b(self.b1).run()
Dense('256').run()
Dropout('0.5').run()
RepeatBegin().b(self.b2).run()
Dense('128', activation='relu').b(self.b3).run()
Dropout('0.2').run()
Dense('128', activation='sigmoid').b(self.b4).run()
Dropout('0.8').run()
Add(input_branch_1=self.b3, input_branch_2=self.b4).b(self.b1).run()
RepeatEnd(self.b2, '2').run()
Dense('10', activation='softmax').run()
SummaryLayer(self.b1).run()
def test_repeat_branch(self):
# branch
Dense('256').b(self.b1).run()
Dropout('0.5').run()
Dense('64').run()
# main Branch
InputLayer(input_shape='784').b(self.b2).run()
Dense('256').run()
Dropout('0.5').run()
Dense('128').run()
RepeatBranch(self.b1, '2').run()
Dense('10', activation='softmax').run()
SummaryLayer(self.b2).run()
def test_relu(self):
InputLayer('784').run()
Dense('512').run()
ReLU(max_value='', negative_slope='0', threshold='0').run()
ReLU(max_value='', negative_slope='0', threshold='0').run()
ReLU(max_value='', negative_slope='0', threshold='0').run()
SummaryLayer().run()
def test_softmax(self):
InputLayer('784').run()
Dense('512').run()
Softmax(axis='-1').run()
Softmax(axis='-1').run()
Softmax(axis='-1').run()
SummaryLayer().run()
def test_thresholded_relu(self):
InputLayer('784').run()
Dense('512').run()
ThresholdedReLU(theta='1.0').run()
ThresholdedReLU(theta='1.0').run()
ThresholdedReLU(theta='1.0').run()
SummaryLayer().run()
def test_elu(self):
InputLayer('784').run()
Dense('512').run()
ELU(alpha='1.0').run()
ELU(alpha='1.0').run()
ELU(alpha='1.0').run()
SummaryLayer().run()
def test_prelu(self):
InputLayer('784').run()
Dense('512').run()
PReLU(alpha_initializer='zeros').run()
PReLU(alpha_initializer='zeros').run()
PReLU(alpha_initializer='zeros').run()
SummaryLayer().run()
def test_leaky_relu(self):
InputLayer('784').run()
Dense('512').run()
LeakyReLU(alpha='0.3').run()
LeakyReLU(alpha='0.3').run()
LeakyReLU(alpha='0.3').run()
SummaryLayer().run()
def test_compile_advanced(self):
# model
SGD(lr='0.25').b(0).run()
InputLayer('784').b(1).run()
Dense('512').run()
CompileAdvanced(0, 'categorical_crossentropy', ['accuracy']).run()
rdd = inputRDD(1)
rdd.show()
print(json.dumps(json.loads(rdd.first().compile_config), indent=4))
print(json.dumps(json.loads(rdd.first().model_config), indent=4))
def test_repeat_branch_multi_branch(self):
# TODO 暂不兼容复用分支快包含多分支类型
# branch
Dense('256').b(self.b1).run()
Dropout('0.5').run()
Dense('64').b(self.b2).run()
Dropout('0.5').run()
Dense('64').b(self.b1).run()
Dropout('0.5').run()
Add(input_branch_1=self.b1, input_branch_2=self.b2).run()
# main Branch
InputLayer(input_shape='784').b(self.b2).run()
Dense('256').run()
Dropout('0.5').run()
Dense('128').run()
RepeatBranch(self.b1, '2').run()
Dense('10', activation='softmax').run()
SummaryLayer(self.b2).run()
def test_compile(self):
InputLayer('784').run()
Dense('512').run()
Compile('categorical_crossentropy', 'rmsprop', ['accuracy']).run()
inputRDD(BRANCH).show()