Metrics in a nutshell

106 words | 1 min read

One of the pillars of machine learning is measuring the performance of models. New tools and techniques to do this are constantly being developed but using multiple tools from different sources can quickly become tedious, inconvenient, and prone to incompatibilities.

The metrics submodule aims to address these issues. It is a collection of many different types of metrics which can be computed on model predictions to evaluate the performance. Whether you are trying to compute regression metrics (such as r2) or classification ones, metrics provides a standard and modular interface for using these metrics and also allows you to add your own!