Put a label on it
PUffy is a Python library designed for creating the best possible models given a dataset with positive-unlabeled data. It implements two-step classification approach. The package is also suitable for considerations on PU classification with fully labeled dataset, as the package presents labeling methods to artificially mimic labeling mechanisms resulting in positive-unlabeled data. Puffy consists of four modules, namely labelling_methods
, rn_methods
, training
and pu_pipeline
.
So far, it offers:
๐ก 4 labeling methods:
-
one method assuming SCAR scenario,
-
three methods being variants of SAR setting,
๐ก 3 techniques for identifying reliable-negatives sample from the unlabeled data:
-
a technique based on KMeans,
-
a technique based on KNN,
-
a technique based on Spy analysis,
-
an iterative classification model for identifying reliable negatives, using Logistic Regression classification,
๐ก visualizations of the model performance,
๐ก a naive classification method for comparison,
๐ก a pipeline, with which presented experiments were conducted.
As of today, the package is meant for binary classification problems.
Full documentation is available here: I'm here!
Thank you for choosing Puffy! This section provides instructions on how to install and set up PUffy on your system.
Before you begin, ensure you have the following prerequisites:
๐ก Python 3.6 or later installed on your system. You can download Python from https://www.python.org/downloads/.
๐ก Ensure pip
(Python package installer) is available. If it's not installed, you can follow the instructions at https://pip.pypa.io/en/stable/installation/.
Visit the GitHub repository to find the latest release of Puffy: https://github.com/kopyta/PUffy
Navigate to the "Releases" section and choose the version you want to install. You can find the releases at: https://github.com/kopyta/PUffy/releases
Download the appropriate package for your platform:
Run the following commands:
wget https://github.com/kopyta/PUffy/releases/v1.0/puffy-1.0-py3-none-any.whl
pip install puffy-1.0-py3-none-any.whl
or
Run the following commands:
wget https://github.com/kopyta/PUffy/releases/v1.0/puffy-1.0.tar.gz
pip install puffy-1.0.tar.gz
You've successfully installed Puffy! Get ready to embark on the exciting challenges of PU data classification. Enjoy exploring the capabilities of Puffy and make the most out of its features! For more information visit our documentation: ๐กCLick๐ก
Happy coding! ๐
The package is a result of the bachelor's thesis, origining from this repository