medcat.components.addons.relation_extraction.pad_seq
====================================================

.. py:module:: medcat.components.addons.relation_extraction.pad_seq


Classes
-------

.. autoapisummary::

   medcat.components.addons.relation_extraction.pad_seq.Pad_Sequence


Module Contents
---------------

.. py:class:: Pad_Sequence(seq_pad_value, label_pad_value = -1)

   .. py:method:: __init__(seq_pad_value, label_pad_value = -1)

      Used in rel_cat.py in RelCAT to create DataLoaders for
      train/test datasets.
      collate_fn for dataloader to collate sequences of different input_ids,
      ent1/ent2, and label lengths into a fixed length batch.
      This is applied per batch and not on the whole DataLoader data,
      padded x sequence, y sequence, x lengths and y lengths of batch.

      :param seq_pad_value: pad value for input_ids.
      :type seq_pad_value: int
      :param label_pad_value: pad value for labels. Defaults to -1.
      :type label_pad_value: int



   .. py:attribute:: seq_pad_value
      :type:  int


   .. py:attribute:: label_pad_value
      :type:  int
      :value: -1



   .. py:method:: __call__(batch)

      Pads a batch of input_ids.

      :param batch: gets the batch of Tensors from
                    RelData.dataset (check __getitem__() method for data returned)
                    and pads the token sequence + labels as needed
                    See https://pytorch.org/docs/stable/_modules/torch/nn/utils/rnn.html#pad_sequence
                    for extra info.
      :type batch: list[torch.Tensor]

      :Returns: **tuple[Tensor, Tensor, Tensor, LongTensor, LongTensor]** -- padded data
                padded input ids, ent1&ent2 start token pos, padded labels,
                padded input_id_lengths, padded labels length