Weakly-Supervised-Learning

This is a package that produces labels using weakly supervised learning with constraint-based methods.

The package contains 2 algorithms, Data Consistent Weak Supervision (DCWS) and Constrained Label Learning (CLL), that are contains code for the following papers

* Constrained Labeling for Weakly Supervised Learning
* Data Consistency for Weakly Supervised Learning

If you use this work in an academic study, please cite our paper

Requirements

The library is tested in Python 3.6 and 3.7.

Its main requirements are Tensorflow and numpy.

Scikit-learn is required to run the experiments.

Examples

We have provided a run_experiment file as an example on both algorithms, along the real datasets. They can all be found under the examples folder.

Logging

Logging is done via TensorBoard. The suggested storage format for each run is by the date/time the expirment was started, and then by dataset, and then by algorithm. Use:

tensorboard --logdir=logs/data_and_time/data_set/algorithm

Example:

tensorboard --logdir=logs/2021_07_28-05:50:52_PM/breast-cancer/CLL

Enjoy!