agarwalmanvi / ml18 Goto Github PK

• create a virtual environment with Conda

conda create --name aif360 python=3.5

To activate the environment, type

conda activate aif360

or source activate aif360 for older version of conda
or activate aif360 for Windows.
To deactivate the environment, type

conda deactivate

or similarly source deactivate, or deactivate

• install requirements provided by requirements.txt.
  This file consists of a list of packages used for this project. You can install the requirements in the virtual environment after activating it by typing the following:

pip install -r requirements.txt

• dataset : a folder which contains the three original data sets used in this project (Adult, Compas, German)
• results : folder which stores all the results in csv format
• codes : original code for the three classifiers and ensemble. There is an output folder which stores the output data after classification

To test the codes, first you need to activate the environment, then go to the codes directory.

source activate aif360
cd codes

We have three scripts which run the project for three different datasets.

• run_adult.py : script to run the project with Adult dataset.
• run_compas.py : script to run the project with Compas dataset.
• run_german.py : script to run the project with German dataset.
  
  To run one of the above files, type:

python run_<adult/compas/german>.py

The three files above save the output (accuracy and fairness scores for multiple runs) to a csv file stored in the results folder.

• analysis-compas.R : script for plotting the results obtained from the previous step.
• alphaCalc.R : script to create the alpha plot which shows combined scores of accuracy and fairness.
  You can also simply run the demo code in jupyter notebook.
• demo.ipynb : script for demo. Store results for accuracy and fairness metrics as a dataframe for all classifiers.