https://pypi.org/project/seqtag/
BiLSTM + CRF for sequence tagging
This is adapted from guillaumegenthial 's original implementation and is made configurable and easy to adapt and use.
Requirements:
This code is tested with all tensorflow versions from 1.3.0 to 1.10.0.
tensorflow is not included in setup.py since, it will remove tensorflow-gpu. Separately install tensorflow by following https://www.tensorflow.org/install/ for this module to work.
Download the 300 dimnesional glove vectors from https://nlp.stanford.edu/projects/glove/
Installation:
The stable version can be installed by running "pip install seqtag"
How to Use:
Create a training directory with train.txt and valid.txt (test.txt is optional) set the config parameters as expected in a configuration file.
An example of the configuration file can be found at https://github.com/bedapudi6788/seqtag/blob/master/example_config.json
Training:
from seqtag import trainer
trainer.train(config_path = 'path_to_config_json')
Running Predictions:
from seqtag import predictor
model = predictor.load_model(path_to_config.json)
predictor.predict(model, ['I', 'am', 'Batman'])
['O', 'O', 'B-PER']
For an usage example take a look at https://github.com/bedapudi6788/Deep-Segmentation/ . seqtag is used for sentence segmentation in this repo.