zipf_classifier 1.2.0

A package to create a deterministic classifier based on Zipf law

Introduction

ZipfClassifier is a classifier that, even though in principle usable on any distribution, leverages the assumption that some kind of datasets su as:

• texts

• images (paper here)

• spoken language (paper here)

follow the Zipf law.

Dependecies

ZipfClassifier uses zipf, another package o’ mine. I also suggest to use dictances for the metrics used in classification.

Installation

pip install zipf_classifier

Working examples and explanation

A jupyter notebook is available with a full explanation, three working examples and respective link to datasets.

License

This package is licensed under MIT license.

FAQs

Frequenctly asked questions down below.

Generally, which metric do you suggest?

Experimental analysis suggests that, in particular when the learning set distributions contain a significant greater number of events than the distribution from the document you are trying to classify, the intersection_squared_hellinger seemed to work best.