chainer_model.py 文件源码

def __call__(self, trainer):
        duration_epoch = time.time() - self.time_epoch
        epoch = trainer.updater.epoch
        loss = trainer.observation["main/loss"].data
        logger.info("epoch: %s, duration: %ds, loss: %.6g.",
                    epoch, duration_epoch, loss)

        # get rnn state
        model = trainer.updater.get_optimizer("main").target
        state = model.predictor.get_state()
        # generate text
        seed = generate_seed(self.text)
        generate_text(model, seed)
        # set rnn back to training state
        model.predictor.set_state(state)

        # reset time
        self.time_epoch = time.time()