pytorch-crf

Description

This package provides an implementation of conditional random field <https://en.wikipedia.org/wiki/Conditional_random_field>_ (CRF) in PyTorch. This implementation borrows mostly from AllenNLP CRF module <https://github.com/allenai/allennlp/blob/master/allennlp/modules/conditional_random_field.py>_ with some modifications.

Requirements

• Python 3.6
• PyTorch 0.4.0

Installation

You can install with pip ::

pip install git+https://github.com/yumoh/torchcrf.git

Examples

In the examples below, we will assume that these lines have been executed

.. code-block:: python

>>> import torch
>>> from torchcrf import CRF
>>> seq_length, batch_size, num_tags = 3, 2, 5
>>> emissions = torch.autograd.Variable(torch.randn(seq_length, batch_size, num_tags), requires_grad=True)
>>> tags = torch.autograd.Variable(torch.LongTensor([[0, 1], [2, 4], [3, 1]]))  # (seq_length, batch_size)
>>> model = CRF(num_tags)

Computing log likelihood

.. code-block:: python

>>> model(emissions, tags)
Variable containing:
-10.0635
[torch.FloatTensor of size 1]

Computing log likelihood with mask

.. code-block:: python

>>> mask = torch.autograd.Variable(torch.ByteTensor([[1, 1], [1, 1], [1, 0]]))  # (seq_length, batch_size)
>>> model(emissions, tags, mask=mask)
Variable containing:
-8.4981
[torch.FloatTensor of size 1]

Decoding

.. code-block:: python

>>> model.decode(emissions)
[[3, 1, 3], [0, 1, 0]]

Decoding with mask

.. code-block:: python

>>> model.decode(emissions, mask=mask)
[[3, 1, 3], [0, 1]]

Installing dependencies

Make sure you setup a virtual environment with Python 3.6 and PyTorch installed. Then, install all the dependencies in requirements.txt file and install this package in development mode. ::

pip install -r requirements.txt
pip install -e .

Running tests

Run pytest in the project root directory.

Running linter

Run flake8 in the project root directory. This will also run mypy, thanks to flake8-mypy package.

.. _LICENSE: https://github.com/kmkurn/pytorch-crf/blob/master/LICENSE.txt