text_field.py 文件源码

def as_tensor(self,
                  padding_lengths: Dict[str, int],
                  cuda_device: int = -1,
                  for_training: bool = True) -> Dict[str, torch.Tensor]:
        tensors = {}
        desired_num_tokens = padding_lengths['num_tokens']
        for indexer_name, indexer in self._token_indexers.items():
            padded_array = indexer.pad_token_sequence(self._indexed_tokens[indexer_name],
                                                      desired_num_tokens, padding_lengths)
            # We use the key of the indexer to recognise what the tensor corresponds to within the
            # field (i.e. the result of word indexing, or the result of character indexing, for
            # example).
            # TODO(mattg): we might someday have a TokenIndexer that needs to use something other
            # than a LongTensor here, and it's not clear how to signal that.  Maybe we'll need to
            # add a class method to TokenIndexer to tell us the type?  But we can worry about that
            # when there's a compelling use case for it.
            tensor = Variable(torch.LongTensor(padded_array), volatile=not for_training)
            tensors[indexer_name] = tensor if cuda_device == -1 else tensor.cuda(cuda_device)
        return tensors