Args: state: A dictionary which corresponds to the output of *LossCompute._make_shard_state(). The values for

Read the config file and pre-/load the models.

Calculate ROUGE scores of sequences passed as an iterator e.g. a list of str, an open file, StringIO or even sys.stdin

Convert subword alignment to word alignment. Args: src (string): tokenized sentence in source language. tgt (string): tokenized s

Translate a batch of sentences.

Manually unload a model. It will free the memory and cancel the timer

Minimally update the indices and frequency of symbol pairs if we merge a pair of symbols, only pairs that overlap with occurrences of this pa

Validate model. valid_iter: validate data iterator Returns: :obj:`nmt.Statistics`: validation loss statistics

Return the length of a video.

Compute the forward loss, possibly in shards in which case this method also runs the backward pass and returns ``None`` as the loss va

Update the internal state of early stopping mechanism, whether to continue training or stop the train procedure. Che

获取对应index的图像，并视情况进行数据增强

Args: report_every(int): Report status every this many sentences start_time(float): manually set report start time. N

A report manager that writes statistics on standard output as well as (optionally) TensorBoard Args: report_ever

Callable class to keep track of early stopping. Args: tolerance(int): number of validation steps without imp

?