Checkpoint may lost when the number of metrics is larger than two

[HugoPu]

Since the numbuer of checkpoint is limited which is defined in model.py as follow:

self.saver = tf.train.Saver(
        tf.global_variables(), max_to_keep=hparams.num_keep_ckpts)

and the number of metrics is not limited which is used in train.py as follow:

if save_on_best and scores[metric] > getattr(hparams, best_metric_label):
        setattr(hparams, best_metric_label, scores[metric])
        model.saver.save(
            sess,
            os.path.join(
                getattr(hparams, best_metric_label + "_dir"), "translate.ckpt"),
            global_step=model.global_step)

If the number of metrics is larger than two, the saved checkpoint may lost. If the number of metrics is larger than num_keep_ckpts, the saved checkpoint certainly lost.