Assert all arguments have the same value

Make an Embeddings instance. Args: opt: the option in current environment. word_dict(Vocab): words dictionary. featur

calculate pairing score :param s: target str :param t: candidate str :return: score, edit distance, length penalty

Compute gradients norm.

Return True if check pass; if check fails and abort is True, raise an Exception, othereise return False.

Various encoder dispatcher function. Args: opt: the option in current environment. embeddings (Embeddings): vocab embeddings

Args: encoder(*Encoder): the various encoder. decoder(*Decoder): the various decoder. multigpu(bool): run

Adds a Test to TestModel according to settings Args: paramSetting: list of tuples of (param, setting) methodname: name of th

Args: src(FloatTensor): a sequence of source tensors with optional feature tensors of size (len x batch).

Various decoder dispatcher function. Args: opt: the option in current environment. embeddings (Embeddings): vocab embeddings

sequence: an iterator returns: the first thing returned by calling next() on the iterator and an iterator created by re-chaining that

Called for each epoch to train.

The encoder hidden is (layers*directions) x batch x dim. We need to convert it to layers x batch x (directions*dim).

Given scores from an expanded dictionary corresponeding to a batch, sums together copies, with a dictionary word when it is a

Compute the loss. Subclass must define this method. Args: batch: the current batch. output: the predict outpu

Called conditionally each epoch to save a snapshot.

This returns user-defined LossCompute object, which is used to compute loss in train/validate process. You can implement your own *LossCo

Make shard state dictionary for shards() to return iterable shards for efficient loss computation. Subclass must define this

path: location of a src or tgt file truncate: maximum sequence length (0 for unlimited) yields: (word, features, nfeat) triples for each

Additional term add to log probability

Tranform the size of the tensors to fit for conv input.

Compute and return a Statistics object. Args: loss(Tensor): the loss computed by the loss criterion. scores(

Adds a Test to TestData according to settings Args: paramSetting: list of tuples of (param, setting) methodname: name of the

Private helper for building standard decoder RNN.

Get an attention mask to avoid using the subsequent info.

Private helper for running the specific RNN forward pass. Must be overriden by all subclasses. Args: input (LongT

Given prob over words for every last beam `wordLk` and attention `attnOut`: Compute and update the beam search. Parameters:

Update when beam advances.