flippers 0.0.1b4

`flippers` is a weak supervision library for creating high quality labels using your domain kownledge and weak supervision sources.

flippers is a Python library for weak supervision, which allows you to leverage your domain knowledge, heuristics and other weak supervision sources to generate high-quality labels for your training data.

Features

flippers includes a number of features for weak supervision, including:

• Simple tools to analyse your labeling functions,
• Multiple label models including a from-scratch implementation of the label model used in the snorkel library and featuring enhanced ways to predict probabilities,
• An extensive documentation with tutorials and an API reference.

Installation

To install the latest version of flippers, simply run:

pip install flippers

Quick Start

Documentation

To quickly get started with flippers, you can begin by exploring the documentation and running through the examples provided. The examples cover a variety of use cases and techniques, which can help you to get a feel for how to apply flippers to your own projects.

Example

• Analyzing your labeling functions:

analysis = flippers.analyis(L_train)

• Training a Label Model and doing inference:

label_model = flippers.models.SnorkelModel(polarities, class_balances)

label_model.fit(L_train)

label_model.predict_proba(L)

Discussion

Troubleshooting

If you have any questions or issues with flippers, please consult the documentation or reach out in the GitHub issues page for support.

Contributing

flippers is an open-source project, and contributions are welcome!

Related projects

• wrench: A weak supervision library created by for benchmarking label models and weak supervision techniques. flippers uses its list of datasets as benchmarks.
• snorkel: A cornerstone weak supervision library created by Snorkel in 2016, now sadly abandonned.