FXTool Kit by PNRIA
About explicability
The project consists in deploying the explicability algorithms from Joao Marques-Silva’s team at ANITI on a web platform.
Indeed they developed multiple algorithms to explain Decision Tree, Random Forest, Naive Bayes, Neural Network … But these algorithms are not accessible to other researchers so the team wants a “showcase site” . The tools are described in recent papers by ANITI’s DeepLever Chair, with prototypes already available for most of the papers.
Please visit Joao Marques-Silva's work on : Github Repo You will see some repos integrated here.
Requirements
Python-3.8.10
$ pip install dash
$ pip install pandas
$ pip install -r requirements.txt
Import graphviz manually
Structure
The structure is adapted for deployment on Heroku, if you don't wish to deploy, you can delete Procfile and runtime.txt.
Running locally
Run app.py then visit localhost
Set the parameters and upload the models (you should upload data for random forest). The instance should either be a .txt at format (feature1=...,feature2=...) or a json file
Deployed app
Visit : App web
Set the parameters and upload the models (you should upload data for random forest). The instance should either be a .txt at format (feature1=...,feature2=...) or a json file
To deploy new changes, please execute this code :
$ git status # view the changes
$ git add . # add all the changes
$ git commit -m 'a description of the changes'
$ git push heroku master
However, for this free deployment, you have a limited slug size. It might need to be deployed somewhere else.
Authors
CNRS-IRIT collaboration with ANITI for PNRIA