def train(self, weights=None):
print("train.py - I am called")
self.trainer.batch_start_time = time.time()
start_global_step = self.trainer.global_step
end_global_step = start_global_step + self.trainer.transition_iters + 1
print("train.py . train . Starting global step: " +
str(start_global_step))
while self.trainer.global_step < end_global_step:
self.trainer.update_transition_value()
self.trainer.dataloader_train.update_next_transition_variable(
self.trainer.transition_variable)
train_iter = iter(self.trainer.dataloader_train)
next_transition_value = utils.compute_transition_value(
self.trainer.global_step + self.trainer.batch_size,
self.trainer.is_transitioning, self.trainer.transition_iters,
self.trainer.latest_switch)
self.trainer.dataloader_train.update_next_transition_variable(
next_transition_value)
#self.trainer.set_weights(weights)
print("train.py . train . while . len(train_iter): " +
str(len(train_iter)))
for i, (real_data, condition, landmarks) in enumerate(train_iter):
self.trainer.logger.update_global_step(
self.trainer.global_step)
if i % 4 == 0:
self.trainer.update_transition_value()
self.trainer.train_step(real_data, condition, landmarks)
# Log data
self.trainer.update_running_average_generator()
self.trainer.maybe_save_validation_checkpoint()
self.trainer.maybe_validate_model()
self.trainer.maybe_save_fake_data(real_data, condition,
landmarks)
self.trainer.global_step += self.trainer.batch_size
print("train.py . train . transition_iters: " +
str(self.trainer.transition_iters))
if self.trainer.global_step >= (self.trainer.latest_switch +
self.trainer.transition_iters):
self.trainer.transition_model()
break
if (i + 1) % 4 == 0:
next_transition_value = utils.compute_transition_value(
self.trainer.global_step + self.trainer.batch_size,
self.trainer.is_transitioning,
self.trainer.transition_iters,
self.trainer.latest_switch)
self.trainer.dataloader_train.update_next_transition_variable(
next_transition_value)
print("End of for loop")
print("train.py . train . Ending global step: " +
str(self.trainer.global_step))
return self.trainer.get_gradients()
def update_transition_value(self):
self.transition_variable = utils.compute_transition_value(
self.global_step, self.is_transitioning, self.transition_iters,
self.latest_switch)
self.discriminator.update_transition_value(self.transition_variable)
self.generator.update_transition_value(self.transition_variable)
self.running_average_generator.update_transition_value(
self.transition_variable)
def train(self):
self.batch_start_time = time.time()
while True:
self.update_transition_value()
self.dataloader_train.update_next_transition_variable(
self.transition_variable)
train_iter = iter(self.dataloader_train)
next_transition_value = utils.compute_transition_value(
self.global_step + self.batch_size, self.is_transitioning, self.transition_iters,
self.latest_switch
)
self.dataloader_train.update_next_transition_variable(
next_transition_value)
for i, (real_data, condition, landmarks) in enumerate(train_iter):
self.logger.update_global_step(self.global_step)
if i % 4 == 0:
self.update_transition_value()
self.train_step(real_data, condition, landmarks)
# Log data
self.update_running_average_generator()
self.maybe_save_validation_checkpoint()
self.maybe_validate_model()
self.maybe_save_fake_data(real_data, condition, landmarks)
self.global_step += self.batch_size
if self.global_step >= (self.latest_switch + self.transition_iters):
self.transition_model()
break
if (i + 1) % 4 == 0:
next_transition_value = utils.compute_transition_value(
self.global_step + self.batch_size, self.is_transitioning, self.transition_iters,
self.latest_switch
)
self.dataloader_train.update_next_transition_variable(
next_transition_value)