Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/Techtonique/teller

Model-agnostic Statistical/Machine Learning explainability (currently Python) for tabular data
https://github.com/Techtonique/teller

explainable-ai xai xai-library

Last synced: 29 days ago
JSON representation

Model-agnostic Statistical/Machine Learning explainability (currently Python) for tabular data

Host: GitHub
URL: https://github.com/Techtonique/teller
Owner: Techtonique
License: bsd-3-clause-clear
Created: 2020-08-14T04:57:59.000Z (over 4 years ago)
Default Branch: master
Last Pushed: 2024-08-07T06:28:44.000Z (6 months ago)
Last Synced: 2025-01-04T20:19:17.761Z (about 1 month ago)
Topics: explainable-ai, xai, xai-library
Language: Jupyter Notebook
Homepage: https://techtonique.github.io/teller/
Size: 5.8 MB
Stars: 8
Watchers: 4
Forks: 1
Open Issues: 0
Metadata Files:
- Readme: README.md
- Changelog: CHANGES.md
- Contributing: CONTRIBUTING.md
- Funding: .github/FUNDING.yml
- License: LICENSE

Awesome Lists containing this project

AwesomeResponsibleAI - teller

README

        ![teller logo](the-teller.png)

  

![PyPI](https://img.shields.io/pypi/v/the-teller) [![PyPI - License](https://img.shields.io/pypi/l/the-teller)](https://github.com/thierrymoudiki/teller/blob/master/LICENSE) [![Downloads](https://pepy.tech/badge/the-teller)](https://pepy.tech/project/the-teller) 

[![HitCount](https://hits.dwyl.com/Techtonique/teller.svg?style=flat-square)](http://hits.dwyl.com/Techtonique/teller)

[![CodeFactor](https://www.codefactor.io/repository/github/techtonique/teller/badge)](https://www.codefactor.io/repository/github/techtonique/teller)

[![Documentation](https://img.shields.io/badge/documentation-is_here-green)](https://techtonique.github.io/teller/)

There is an increasing need for __transparency__ and __fairness__ in Machine Learning (ML) models  predictions. Consider for example a banker who has to explain to a client why his/her loan application is rejected, or a healthcare professional who must explain what constitutes his/her diagnosis. Some ML models are indeed very accurate, but are considered to be hard to explain, relatively to popular linear models. 

__Source of figure__: James, Gareth, et al. An introduction to statistical learning. Vol. 112. New York: springer, 2013.

![Source: James, Gareth, et al. An introduction to statistical learning. Vol. 112. New York: springer, 2013.](image1.png)

We do not want to sacrifice this high accuracy to explainability.  Hence: __ML explainability__. There are a lot of ML explainability tools out there, _in the wild_.

The `teller` is a __model-agnostic tool for ML explainability__. _Agnostic_, as long as the input ML model possesses methods `fit` and `predict`, and is applied to tabular data. The `teller` relies on:

- [Finite differences](https://en.wikipedia.org/wiki/Finite_difference) to explain ML models predictions: a little increase in model's explanatory variables + a little decrease, and we can obtain approximate sensitivities of its predictions to changes in these explanatory variables. 

- [Conformal prediction](https://en.wikipedia.org/wiki/Conformal_prediction) (so far, as of october 2022) to obtain prediction intervals for ML methods

## Installation 

- From Pypi, stable version:

```bash

pip install the-teller

```

- From Github, for the development version: 

```bash

pip install git+https://github.com/Techtonique/teller.git

```

## Package description

These notebooks will be some good introductions:

- [Heterogeneity of marginal effects](/teller/demo/thierrymoudiki_011119_boston_housing.ipynb)

- [Significance of marginal effects](/teller/demo/thierrymoudiki_081119_boston_housing.ipynb)

- [Model comparison](/teller/demo/thierrymoudiki_151119_boston_housing.ipynb)

- [Classification](/teller/demo/thierrymoudiki_041219_breast_cancer_classif.ipynb)

- [Interactions](/teller/demo/thierrymoudiki_041219_boston_housing_interactions.ipynb)

- [Prediction intervals for regression](/teller/demo/thierrymoudiki_031022_diabetes_pred_interval.ipynb)

## Contributing

Your contributions are welcome, and valuable. Please, make sure to __read__ the [Code of Conduct](CONTRIBUTING.md) first. 

If you're not comfortable with Git/Version Control yet, please use [this form](https://forms.gle/Y18xaEHL78Fvci7r8).

In Pull Requests, let's strive to use [`black`](https://black.readthedocs.io/en/stable/) for formatting: 

```bash

pip install black

black --line-length=80 file_submitted_for_pr.py

```

## API Documentation

[https://techtonique.github.io/teller/](https://techtonique.github.io/teller/)

## Dependencies 

- Numpy

- Pandas

- Scipy

- scikit-learn

## Citation

```

@misc{moudiki2019teller,

	author={Moudiki, T.},

	title={\code{teller}, {M}odel-agnostic {M}achine {L}earning explainability},

	howpublished={\url{https://github.com/thierrymoudiki/teller}},

	note={BSD 3-Clause Clear License. Version 0.x.x.},

	year={2019--2020}

	}

```

## References

For **sensitivity analysis**: 

- Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. (1992). Numerical recipes in C (Vol. 2). Cambridge: Cambridge university press.

- Jones E, Oliphant E, Peterson P, et al. SciPy: Open Source Scientific Tools for Python, 2001-, http://www.scipy.org/ [Online; accessed 2019-01-04]

- Scikit-learn: Machine Learning in Python, Pedregosa et al., JMLR 12, pp. 2825-2830, 2011.

For **prediction intervals**: 

- Romano, Y., Patterson, E., & Candes, E. (2019). Conformalized quantile regression. Advances in neural information processing systems, 32.

## License

[BSD 3-Clause](LICENSE) © Thierry Moudiki, 2019.