def train(self):
"""
Function to train our model. Calls train_epoch function every epoch.
Also performs learning rate annhealing
"""
logger.info("train_loader length: {}".format(len(self.train_loader)))
logger.info("target_loader length: {}".format(len(self.target_loader)))
iters_per_epoch = min(len(self.target_loader), len(self.train_loader))
self.iters_per_epoch = iters_per_epoch - (iters_per_epoch %
self.batch_size) - 1
logger.info("iters_per_epoch :{}".format(self.iters_per_epoch))
self.max_epoch = args.max_epoch
for epoch in tqdm.trange(self.epoch, args.max_epoch, desc='Train'):
self.epoch = epoch
self.optim = step_scheduler(self.optim, self.epoch,
base_lr_schedule, "base model")
self.optimD = step_scheduler(self.optimD, self.epoch,
dis_lr_schedule,
"discriminater model")
self.model.train()
self.netD.train()
self.train_epoch()
self.model.eval()
self.validate()
self.model.train() # return to training mode
def train(self):
"""
Function to train our model. Calls train_epoch function every epoch.
Also performs learning rate annhealing
"""
logger.info("train_loader length: {}".format(len(self.train_loader)))
logger.info("target_loader length: {}".format(len(self.target_loader)))
iters_per_epoch = min(len(self.target_loader), len(self.train_loader))
self.iters_per_epoch = iters_per_epoch - (iters_per_epoch %
self.batch_size) - 1
logger.info("iters_per_epoch :{}".format(self.iters_per_epoch))
max_epoch = int(math.ceil(self.max_iter / self.iters_per_epoch))
self.max_epoch = max_epoch
for epoch in tqdm.trange(self.epoch, max_epoch, desc='Train'):
self.epoch = epoch
if self.epoch % 8 == 0 and self.epoch > 0:
self.optim = step_scheduler(self.optim, self.epoch)
self.train_epoch()
if self.iteration >= self.max_iter:
break
