https://github.com/iesl/dilated-cnn-ner

Dilated CNNs for NER in TensorFlow
https://github.com/iesl/dilated-cnn-ner

cnns machine-learning named-entity-recognition natural-language-processing neural-networks tensorflow

Last synced: 3 days ago
JSON representation

Dilated CNNs for NER in TensorFlow

• Host: GitHub
• URL: https://github.com/iesl/dilated-cnn-ner
• Owner: iesl
• Created: 2017-04-07T07:44:11.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2019-03-09T01:04:42.000Z (almost 6 years ago)
• Last Synced: 2025-01-03T03:09:11.599Z (11 days ago)
• Topics: cnns, machine-learning, named-entity-recognition, natural-language-processing, neural-networks, tensorflow
• Language: Python
• Size: 160 KB
• Stars: 243
• Watchers: 23
• Forks: 60
• Open Issues: 7
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# dilated-cnn-ner

This code implements the models described in the paper

"[Fast and Accurate Entity Recognition with Iterated Dilated Convolutions](https://arxiv.org/abs/1702.02098)"

by [Emma Strubell](https://cs.umass.edu/~strubell), [Patrick Verga](https://cs.umass.edu/~pat),

[David Belanger](https://cs.umass.edu/~belanger) and [Andrew McCallum](https://cs.umass.edu/~mccallum).

Requirements

-----

This code uses TensorFlow v[1.0, 1.4) and Python 2.7.

It will probably train on a CPU, but honestly we haven't tried, and highly recommend training on a GPU.

Setup

-----

1. Set up environment variables. For example, from the root directory of this project:

  ```

  export DILATED_CNN_NER_ROOT=`pwd`

  export DATA_DIR=/path/to/conll-2003

  ```

2. Get some pretrained word embeddings, e.g. [SENNA embeddings](http://ronan.collobert.com/senna/download.html) or

  [Glove embeddings](https://nlp.stanford.edu/projects/glove/). The code expects a space-separated file

  with one word and its embedding per line, e.g.:

   ```

   word 0.45 0.67 0.99 ...

   ```

   Make a directory for the embeddings:

   ```

   mkdir -p data/embeddings

   ```

   and place the file there.

3. Perform all data preprocessing for a given configuration. For example:

  ```

  ./bin/preprocess.sh conf/conll/dilated-cnn.conf

  ```

  This calls `preprocess.py`, which loads the data from text files, maps the tokens, labels and any other features to

  integers, and writes to TensorFlow tfrecords.

Training

----

Once the data preprocessing is completed, you can train a tagger:

  ```

  ./bin/train-cnn.sh conf/conll/dilated-cnn.conf

  ```

Evaluation

----

By default, the trainer will write the model which achieved the best dev F1. To evaluate a saved model on the dev set:

  ```

  ./bin/eval-cnn.sh conf/conll/dilated-cnn.conf --load_model path/to/model

  ```

To evaluate a saved model on the test set:

  ```

  ./bin/eval-cnn.sh conf/conll/dilated-cnn.conf test --load_model path/to/model

  ```

Configs

----

Configuration files (`conf/*`) specify all the data, parameters, etc. for an experiment.