simple-co-train 1.0.0

A simple co-training library built on Keras.

A simple co-training library built with Keras.

• Free software: MIT license

Installation

Before installing simple-co-train, please install one of Keras’ backend engines: TensorFlow, Theano, or CNTK.

pip install simple-co-train

Documentation

Basic usage:

from sctrain import CoTrainer, SelectionStrategy
from sctrain.results import print_results

trainer = CoTrainer(
    data_path='imdb.csv',  # this can be a directory, e.g. 'data'
    x_name='review',  # optional, defaults to 'text'
    y_name='sentiment',  # optional, defaults to 'label'
    unlabelled_size=0.9, # optional, what portion of total data should be used as unlabelled
    train_size=0.8, # optional, what portion of labelled data should be used as training data
    mapping={'negative': 0, 'positive': 1}  # optional mapping, y column must be 0 or 1
    selection = SelectionStrategy.UNSURE_ONLY # optional, can be CONFIDENT_ONLY or BOTH
)
# run the co-training, this may take a while...
trainer.run()
# print out accuracy, f1 score, precision, recall, and labelled samples at each co-training round
print_results(trainer)

Development

To run the all tests run:

tox

Note, to combine the coverage data from all the tox environments run:

Windows	set PYTEST_ADDOPTS=--cov-append tox
Other	PYTEST_ADDOPTS=--cov-append tox

Changelog

0.1.0 (2020-05-07)

• First release on PyPI.