https://github.com/Quantmetry/resources-intelligibility

Some resources for intelligibility analysis of machine learning models.
https://github.com/Quantmetry/resources-intelligibility

Last synced: 2 months ago
JSON representation

Some resources for intelligibility analysis of machine learning models.

• Host: GitHub
• URL: https://github.com/Quantmetry/resources-intelligibility
• Owner: Quantmetry
• License: mit
• Created: 2018-11-20T16:35:36.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2018-11-22T11:05:34.000Z (about 6 years ago)
• Last Synced: 2024-08-02T02:07:45.848Z (6 months ago)
• Language: Jupyter Notebook
• Size: 687 KB
• Stars: 4
• Watchers: 4
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome_quantmetry - Quantmetry intelligibility resources

README

        
# resources-intelligibility

Some resources for intelligibility analysis of machine learning models, mostly in French.

Notes on setting up the project

-------------------------------

- with a version of *python3* installed (tested with python 3.6), make sure you have access to *pip*.

- with the below instructions, create a local virtual environnment and activate it

- install requirements.txt

  ```

  $ python3 -m venv .venv

  $ source .venv/bin/activate

  (.venv) $ pip install -r requirements.txt

  ```

- Go to the *data/* folder and download (~0.5Mo) the required data with the link you can find in *data/howtogetdata.txt*. At the end of this step, you should have a *carInsurance_train.csv* file in the *data/* folder.

- Start a jupyter server.

  ```

  (.venv) $ jupyter notebook

  ```

Features

--------

In the *notebooks/* folder, you will find some demos of several intelligibility techniques:

- Partie1\_Construction\_Modèle.ipynb

- Partie2\_Analyse_sensibilité\_des\_prédictions.ipynb

- Partie3\_Décomposition\_en\_contributions.ipynb

- Partie4\_Décomposition\_en\_règles.ipynb

You should run *Partie1* first because it will write a pickle with data and model, used by other notebooks. Afterwards, notebooks are independant.

Credits

-------

This work has been done by Quantmetry R&D, 2018.