Python Tensor.new_zeros方法代码示例

# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import new_zeros [as 别名]
    def get_best_span(span_start_logits: torch.Tensor, span_end_logits: torch.Tensor) -> torch.Tensor:
        if span_start_logits.dim() != 2 or span_end_logits.dim() != 2:
            raise ValueError("Input shapes must be (batch_size, passage_length)")
        batch_size, passage_length = span_start_logits.size()
        max_span_log_prob = [-1e20] * batch_size
        span_start_argmax = [0] * batch_size
        best_word_span = span_start_logits.new_zeros((batch_size, 2), dtype=torch.long)

        span_start_logits = span_start_logits.detach().cpu().numpy()
        span_end_logits = span_end_logits.detach().cpu().numpy()

        for b in range(batch_size):  # pylint: disable=invalid-name
            for j in range(passage_length):
                val1 = span_start_logits[b, span_start_argmax[b]]
                if val1 < span_start_logits[b, j]:
                    span_start_argmax[b] = j
                    val1 = span_start_logits[b, j]

                val2 = span_end_logits[b, j]

                if val1 + val2 > max_span_log_prob[b]:
                    best_word_span[b, 0] = span_start_argmax[b]
                    best_word_span[b, 1] = j
                    max_span_log_prob[b] = val1 + val2
        return best_word_span

# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import new_zeros [as 别名]
def add_sentence_boundary_token_ids(tensor: torch.Tensor,
                                    mask: torch.Tensor,
                                    sentence_begin_token: Any,
                                    sentence_end_token: Any) -> Tuple[torch.Tensor, torch.Tensor]:
    """
    Add begin/end of sentence tokens to the batch of sentences.
    Given a batch of sentences with size ``(batch_size, timesteps)`` or
    ``(batch_size, timesteps, dim)`` this returns a tensor of shape
    ``(batch_size, timesteps + 2)`` or ``(batch_size, timesteps + 2, dim)`` respectively.

    Returns both the new tensor and updated mask.

    Parameters
    ----------
    tensor : ``torch.Tensor``
        A tensor of shape ``(batch_size, timesteps)`` or ``(batch_size, timesteps, dim)``
    mask : ``torch.Tensor``
         A tensor of shape ``(batch_size, timesteps)``
    sentence_begin_token: Any (anything that can be broadcast in torch for assignment)
        For 2D input, a scalar with the <S> id. For 3D input, a tensor with length dim.
    sentence_end_token: Any (anything that can be broadcast in torch for assignment)
        For 2D input, a scalar with the </S> id. For 3D input, a tensor with length dim.

    Returns
    -------
    tensor_with_boundary_tokens : ``torch.Tensor``
        The tensor with the appended and prepended boundary tokens. If the input was 2D,
        it has shape (batch_size, timesteps + 2) and if the input was 3D, it has shape
        (batch_size, timesteps + 2, dim).
    new_mask : ``torch.Tensor``
        The new mask for the tensor, taking into account the appended tokens
        marking the beginning and end of the sentence.
    """
    # TODO: matthewp, profile this transfer
    sequence_lengths = mask.sum(dim=1).detach().cpu().numpy()
    tensor_shape = list(tensor.data.shape)
    new_shape = list(tensor_shape)
    new_shape[1] = tensor_shape[1] + 2
    tensor_with_boundary_tokens = tensor.new_zeros(*new_shape)
    if len(tensor_shape) == 2:
        tensor_with_boundary_tokens[:, 1:-1] = tensor
        tensor_with_boundary_tokens[:, 0] = sentence_begin_token
        for i, j in enumerate(sequence_lengths):
            tensor_with_boundary_tokens[i, j + 1] = sentence_end_token
        new_mask = (tensor_with_boundary_tokens != 0).long()
    elif len(tensor_shape) == 3:
        tensor_with_boundary_tokens[:, 1:-1, :] = tensor
        for i, j in enumerate(sequence_lengths):
            tensor_with_boundary_tokens[i, 0, :] = sentence_begin_token
            tensor_with_boundary_tokens[i, j + 1, :] = sentence_end_token
        new_mask = ((tensor_with_boundary_tokens > 0).long().sum(dim=-1) > 0).long()
    else:
        raise ValueError("add_sentence_boundary_token_ids only accepts 2D and 3D input")

    return tensor_with_boundary_tokens, new_mask

# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import new_zeros [as 别名]
def remove_sentence_boundaries(tensor: torch.Tensor,
                               mask: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
    """
    Remove begin/end of sentence embeddings from the batch of sentences.
    Given a batch of sentences with size ``(batch_size, timesteps, dim)``
    this returns a tensor of shape ``(batch_size, timesteps - 2, dim)`` after removing
    the beginning and end sentence markers.  The sentences are assumed to be padded on the right,
    with the beginning of each sentence assumed to occur at index 0 (i.e., ``mask[:, 0]`` is assumed
    to be 1).

    Returns both the new tensor and updated mask.

    This function is the inverse of ``add_sentence_boundary_token_ids``.

    Parameters
    ----------
    tensor : ``torch.Tensor``
        A tensor of shape ``(batch_size, timesteps, dim)``
    mask : ``torch.Tensor``
         A tensor of shape ``(batch_size, timesteps)``

    Returns
    -------
    tensor_without_boundary_tokens : ``torch.Tensor``
        The tensor after removing the boundary tokens of shape ``(batch_size, timesteps - 2, dim)``
    new_mask : ``torch.Tensor``
        The new mask for the tensor of shape ``(batch_size, timesteps - 2)``.
    """
    # TODO: matthewp, profile this transfer
    sequence_lengths = mask.sum(dim=1).detach().cpu().numpy()
    tensor_shape = list(tensor.data.shape)
    new_shape = list(tensor_shape)
    new_shape[1] = tensor_shape[1] - 2
    tensor_without_boundary_tokens = tensor.new_zeros(*new_shape)
    new_mask = tensor.new_zeros((new_shape[0], new_shape[1]), dtype=torch.long)
    for i, j in enumerate(sequence_lengths):
        if j > 2:
            tensor_without_boundary_tokens[i, :(j - 2), :] = tensor[i, 1:(j - 1), :]
            new_mask[i, :(j - 2)] = 1

    return tensor_without_boundary_tokens, new_mask

# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import new_zeros [as 别名]
 def _get_target_token_embeddings(self,
                                  token_embeddings: torch.Tensor,
                                  mask: torch.Tensor,
                                  direction: int) -> torch.Tensor:
     # Need to shift the mask in the correct direction
     zero_col = token_embeddings.new_zeros(mask.size(0), 1).byte()
     if direction == 0:
         # forward direction, get token to right
         shifted_mask = torch.cat([zero_col, mask[:, 0:-1]], dim=1)
     else:
         shifted_mask = torch.cat([mask[:, 1:], zero_col], dim=1)
     return token_embeddings.masked_select(shifted_mask.unsqueeze(-1)).view(-1, self._forward_dim)

# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import new_zeros [as 别名]
    def _get_initial_state(self,
                           encoder_outputs: torch.Tensor,
                           mask: torch.Tensor,
                           actions: List[List[ProductionRule]]) -> GrammarBasedState:

        batch_size = encoder_outputs.size(0)
        # This will be our initial hidden state and memory cell for the decoder LSTM.
        final_encoder_output = util.get_final_encoder_states(encoder_outputs,
                                                             mask,
                                                             self._encoder.is_bidirectional())
        memory_cell = encoder_outputs.new_zeros(batch_size, self._encoder.get_output_dim())
        initial_score = encoder_outputs.data.new_zeros(batch_size)

        # To make grouping states together in the decoder easier, we convert the batch dimension in
        # all of our tensors into an outer list.  For instance, the encoder outputs have shape
        # `(batch_size, utterance_length, encoder_output_dim)`.  We need to convert this into a list
        # of `batch_size` tensors, each of shape `(utterance_length, encoder_output_dim)`.  Then we
        # won't have to do any index selects, or anything, we'll just do some `torch.cat()`s.
        initial_score_list = [initial_score[i] for i in range(batch_size)]
        encoder_output_list = [encoder_outputs[i] for i in range(batch_size)]
        utterance_mask_list = [mask[i] for i in range(batch_size)]
        initial_rnn_state = []
        for i in range(batch_size):
            initial_rnn_state.append(RnnStatelet(final_encoder_output[i],
                                                 memory_cell[i],
                                                 self._first_action_embedding,
                                                 self._first_attended_utterance,
                                                 encoder_output_list,
                                                 utterance_mask_list))

        initial_grammar_state = [self._create_grammar_state(actions[i]) for i in range(batch_size)]

        initial_state = GrammarBasedState(batch_indices=list(range(batch_size)),
                                          action_history=[[] for _ in range(batch_size)],
                                          score=initial_score_list,
                                          rnn_state=initial_rnn_state,
                                          grammar_state=initial_grammar_state,
                                          possible_actions=actions,
                                          debug_info=None)
        return initial_state

# 需要导入模块: from torch import Tensor [as 别名]
# 或者: from torch.Tensor import new_zeros [as 别名]
    def _get_best_span_yesno_followup(span_start_logits: torch.Tensor,
                                      span_end_logits: torch.Tensor,
                                      span_yesno_logits: torch.Tensor,
                                      span_followup_logits: torch.Tensor,
                                      max_span_length: int) -> torch.Tensor:
        # Returns the index of highest-scoring span that is not longer than 30 tokens, as well as
        # yesno prediction bit and followup prediction bit from the predicted span end token.
        if span_start_logits.dim() != 2 or span_end_logits.dim() != 2:
            raise ValueError("Input shapes must be (batch_size, passage_length)")
        batch_size, passage_length = span_start_logits.size()
        max_span_log_prob = [-1e20] * batch_size
        span_start_argmax = [0] * batch_size

        best_word_span = span_start_logits.new_zeros((batch_size, 4), dtype=torch.long)

        span_start_logits = span_start_logits.data.cpu().numpy()
        span_end_logits = span_end_logits.data.cpu().numpy()
        span_yesno_logits = span_yesno_logits.data.cpu().numpy()
        span_followup_logits = span_followup_logits.data.cpu().numpy()
        for b_i in range(batch_size):  # pylint: disable=invalid-name
            for j in range(passage_length):
                val1 = span_start_logits[b_i, span_start_argmax[b_i]]
                if val1 < span_start_logits[b_i, j]:
                    span_start_argmax[b_i] = j
                    val1 = span_start_logits[b_i, j]
                val2 = span_end_logits[b_i, j]
                if val1 + val2 > max_span_log_prob[b_i]:
                    if j - span_start_argmax[b_i] > max_span_length:
                        continue
                    best_word_span[b_i, 0] = span_start_argmax[b_i]
                    best_word_span[b_i, 1] = j
                    max_span_log_prob[b_i] = val1 + val2
        for b_i in range(batch_size):
            j = best_word_span[b_i, 1]
            yesno_pred = np.argmax(span_yesno_logits[b_i, j])
            followup_pred = np.argmax(span_followup_logits[b_i, j])
            best_word_span[b_i, 2] = int(yesno_pred)
            best_word_span[b_i, 3] = int(followup_pred)
        return best_word_span