# Explainable imbalanceD learninG compARatOr
[](https://github.com/adrianstando/EDGAR/actions/workflows/main-check.yaml)
## Overview
The usage of many balancing methods like Random Undersampling, Random Oversampling, SMOTE, NearMiss is a very popular
solution when dealing with imbalanced data. However, a question can be posed of whether these techniques can change the
model behaviour or the relationships present in data.
As there are many kinds of Machine Learning models, this package provides model-agnostic tools to investigate the model
behaviour and its changes. These tools are also known as Explainable Artificial Intelligence (XAI) tools and include
techniques such as Partial Dependence Profile (PDP), Accumulated Local Effects (ALE) and Variable Importance (VI).
Apart from that, the package implements novel methods to compare the explanations, which are *Standard Deviation of
Distances* (for PDP and ALE) and the Wilcoxon statistical test (for VI).
Generally speaking, this package aims to giving a user-friendly interface to investigate whether the described
phenomena take place.
The package was written in Python and consists of four modules: *dataset*, *balancing*, *model* and *explain*.
It provides a simple and user-friendly interface which aims to automate the process of data balancing with different
methods, training Machine Learning models and calculating PDP/ALE/VI explanations. The package can be used for one input
dataset or for a number of datasets arranged in arrays or nested arrays.
## Technologies
The package was written in Python and was checked to be compatible with Python 3.8, Python 3.9 and Python 3.10.
It uses most popular libraries for Machine Learning in Python:
* pandas, NumPy
* scikit-learn, xgboost
* imbalanced-learn
* dalex
* scipy, statsmodels
* matplotlib
* openml
## User Manual
User Manual is available as a part of the documentation, [here](https://adrianstando.github.io/edgaro/source/manual.html)
## Installation
The `edgaro` package is available on [PyPI](https://pypi.org/project/edgaro/) and can be installed by:
```console
pip install edgaro
```
## Documentation
The documentation is available at [adrianstando.github.io/edgaro](https://adrianstando.github.io/edgaro)
## Project purpose
This package was created for the purpose of my Engineering Thesis *"The impact of data balancing on model behaviour with
Explainable Artificial Intelligence tools in imbalanced classification problems"*.
Raw data
{
"_id": null,
"home_page": "https://github.com/adrianstando/edgaro",
"name": "edgaro",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.8, <4",
"maintainer_email": "",
"keywords": "XAI,imbalance,machine learning,AI",
"author": "Adrian Sta\u0144do",
"author_email": "adrian.j.stando@gmail.com",
"download_url": "https://files.pythonhosted.org/packages/09/f4/ce4db16126edd931b8148f5c8e3b11db52dbb0995ada053cb14f4815e91e/edgaro-1.0.2.1.tar.gz",
"platform": null,
"description": "# Explainable imbalanceD learninG compARatOr\n\n[](https://github.com/adrianstando/EDGAR/actions/workflows/main-check.yaml)\n\n## Overview\n\nThe usage of many balancing methods like Random Undersampling, Random Oversampling, SMOTE, NearMiss is a very popular \nsolution when dealing with imbalanced data. However, a question can be posed of whether these techniques can change the\nmodel behaviour or the relationships present in data. \n\nAs there are many kinds of Machine Learning models, this package provides model-agnostic tools to investigate the model\nbehaviour and its changes. These tools are also known as Explainable Artificial Intelligence (XAI) tools and include\ntechniques such as Partial Dependence Profile (PDP), Accumulated Local Effects (ALE) and Variable Importance (VI). \n\nApart from that, the package implements novel methods to compare the explanations, which are *Standard Deviation of \nDistances* (for PDP and ALE) and the Wilcoxon statistical test (for VI).\n\nGenerally speaking, this package aims to giving a user-friendly interface to investigate whether the described \nphenomena take place.\n\nThe package was written in Python and consists of four modules: *dataset*, *balancing*, *model* and *explain*. \nIt provides a simple and user-friendly interface which aims to automate the process of data balancing with different \nmethods, training Machine Learning models and calculating PDP/ALE/VI explanations. The package can be used for one input\ndataset or for a number of datasets arranged in arrays or nested arrays.\n\n## Technologies\n\nThe package was written in Python and was checked to be compatible with Python 3.8, Python 3.9 and Python 3.10.\n\nIt uses most popular libraries for Machine Learning in Python:\n\n* pandas, NumPy\n* scikit-learn, xgboost\n* imbalanced-learn\n* dalex\n* scipy, statsmodels\n* matplotlib\n* openml\n\n## User Manual\n\nUser Manual is available as a part of the documentation, [here](https://adrianstando.github.io/edgaro/source/manual.html)\n\n## Installation\n\nThe `edgaro` package is available on [PyPI](https://pypi.org/project/edgaro/) and can be installed by:\n\n```console\npip install edgaro\n```\n\n## Documentation\n\nThe documentation is available at [adrianstando.github.io/edgaro](https://adrianstando.github.io/edgaro)\n\n## Project purpose\n\nThis package was created for the purpose of my Engineering Thesis *\"The impact of data balancing on model behaviour with \nExplainable Artificial Intelligence tools in imbalanced classification problems\"*.\n",
"bugtrack_url": null,
"license": "",
"summary": "Explainable imbalanceD learninG compARatOr",
"version": "1.0.2.1",
"project_urls": {
"Code repository": "https://github.com/adrianstando/edgaro",
"Documentation": "https://adrianstando.github.io/edgaro",
"Homepage": "https://github.com/adrianstando/edgaro"
},
"split_keywords": [
"xai",
"imbalance",
"machine learning",
"ai"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "023cc07d1d80dd5b53e3161e8f61b8e4767812282d19161a0958b88168853ed2",
"md5": "55414237f22ec3ba0d79ca6d8aecf70f",
"sha256": "ce93110987f93c0cce33f1de91a5b3a44bcd3c545de9b0eb42c6349a367c1a4e"
},
"downloads": -1,
"filename": "edgaro-1.0.2.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "55414237f22ec3ba0d79ca6d8aecf70f",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8, <4",
"size": 60831,
"upload_time": "2023-06-16T20:12:53",
"upload_time_iso_8601": "2023-06-16T20:12:53.028377Z",
"url": "https://files.pythonhosted.org/packages/02/3c/c07d1d80dd5b53e3161e8f61b8e4767812282d19161a0958b88168853ed2/edgaro-1.0.2.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "09f4ce4db16126edd931b8148f5c8e3b11db52dbb0995ada053cb14f4815e91e",
"md5": "292e374d75376f84de0ed8e8444a8e8c",
"sha256": "4ef296924b34eb06ccea89bf606571f6f7d5b8d692432b7416b563260d54bb6f"
},
"downloads": -1,
"filename": "edgaro-1.0.2.1.tar.gz",
"has_sig": false,
"md5_digest": "292e374d75376f84de0ed8e8444a8e8c",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8, <4",
"size": 61613,
"upload_time": "2023-06-16T20:12:55",
"upload_time_iso_8601": "2023-06-16T20:12:55.707763Z",
"url": "https://files.pythonhosted.org/packages/09/f4/ce4db16126edd931b8148f5c8e3b11db52dbb0995ada053cb14f4815e91e/edgaro-1.0.2.1.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-06-16 20:12:55",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "adrianstando",
"github_project": "edgaro",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"requirements": [],
"tox": true,
"lcname": "edgaro"
}
JSON
