class SupervisedManager(object):
"""docstring for SupervisedManager."""
def __init__(self, experimentname=None):
# class attributes
self._i = 0
self.experimentname = experimentname
self.model = None
self.weightpath = None
self.loss = None
self.callbacks = []
def setWeights(self, model, weightpath=None, multi_gpu=False):
self.model = model
self.weightpath = weightpath
if weightpath:
print("[*] Loading weights...")
self.model.load_weights(weightpath, by_name=True)
print("Done\n")
def setModel(self, modelpath=None):
self.model = load_model(modelpath)
def setFreeze(self, freeze):
self.freeze = freeze
for layer in self.model.layers:
layer.trainable = not layer.name in freeze
def setLoss(self, loss):
self.loss = loss
def setOptimizer(self, optimizer):
self.optimizer = optimizer
def compile(self):
self.model.compile(optimizer=self.optimizer, loss=self.loss)
def setGroundTruth(self, datapath, ratio_valid=0.1):
# PATH = "../resources"
self.gt = GroundTruth(path=datapath, split=ratio_valid)
def setInputGenerator(self, nb_timesteps, batch_size):
self.batch_size = batch_size
self.generator = InputGenerator(self.gt,
chunk_size=nb_timesteps,
batch_size=batch_size)
def setCallbacks(self, callbacks):
self.callbacks.append(callbacks)
def train(self, epochs=15, initial_epoch=0):
history = self.model.fit_generator(
self.generator.generate(set="train"),
steps_per_epoch=self.generator.steps_per_epoch_train,
validation_data=self.generator.generate(set="valid"),
validation_steps=self.generator.steps_per_epoch_valid,
epochs=epochs,
callbacks=self.callbacks,
max_queue_size=10,
workers=1,
verbose=1,
initial_epoch=initial_epoch,
)
return history
