README.md

In mstaniak/egalitaRian: Compare Validation and Training Datasets and Explore Differences

egalitaRian

Installation

To get started, install the development version:

devtools::install_github("mstaniak/egalitaRian")

See the latest changes.

Overview

egalitaRian is an R package which helps addresses the biased training problem by providing several validation tools. After data are preprocessed using egalitarian function, several explainers can be used:

(Already implemented) * compare_distributions with its plot method helps compare distributions of a single variable in training and validation (test) datasets, * get_distribution_distances provides statistics that describe distance between distributions of each variable in training and validation datasets. Currently implemented: Kolmogorov-Smirnov distance and Chi-squared distance. * local_explanation leverages local explanation methods such as LIME, LIVE and breakDown to find problematic variables which contributed to misclassification of a single observation and overlays values of these variables on their distributions in training dataset (plot method). Currently, this method is implemented only for classification problems. * global_similarity uses heuristic method based on k means clustering to assess similarity of training and validation datasets.

mstaniak/egalitaRian documentation built on Aug. 26, 2019, 11:11 p.m.