def process(self, global_properties={}, properties={}, container={}):
model = container[properties["model"]]
optimization_def = properties["optimization"]
if optimization_def["algorithm"] == "StochasticGradientDescent":
optimizer = SGD()
if "nesterov" in optimization_def:
optimizer.nesterov = optimization_def["nesterov"]
if optimization_def["algorithm"] == "RMSprop":
optimizer = RMSprop()
if optimization_def["algorithm"] == "Adam":
optimizer = Adam()
metrics = properties["metrics"].split(",")
if properties["loss"] == "CategoricalCrossEntropy":
loss = "categorical_crossentropy"
model.compile(loss=loss, optimizer=optimizer, metrics=metrics)
if optimization_def["algorithm"] == "StochasticGradientDescent":
if "learning_rate" in optimization_def:
K.set_value(model.optimizer.lr,
optimization_def["learning_rate"])
if "momentum" in optimization_def:
K.set_value(model.optimizer.momentum,
optimization_def["momentum"])
if "decay" in optimization_def:
K.set_value(
model.optimizer.decay,
optimization_def["decay"] / global_properties["epochs"])
if optimization_def["algorithm"] == "RMSprop":
if "learning_rate" in optimization_def:
K.set_value(model.optimizer.lr,
optimization_def["learning_rate"])
if optimization_def["algorithm"] == "Adam":
if "learning_rate" in optimization_def:
K.set_value(model.optimizer.lr,
optimization_def["learning_rate"])
container[properties["model"]] = model
return container
