nrkarthikeyan

Running the HELOC notebook produces this error: `module 'torch.jit' has no attribute '_script_if_tracing' ` which is resolved by using a different version of pytorch `!pip install torch==1.6.0+cpu torchvision==0.7.0+cpu -f https://download.pytorch.org/whl/torch_stable.html`...

Allow installation of specific algorithms and migrate to workflows from travis for CI as in https://github.com/Trusted-AI/AIF360

Corrected german credit data

3

The widely used german credit data (that is already available in the toolkit) apparently has coding errors, so consider including https://archive.ics.uci.edu/ml/datasets/South+German+Credit+%28UPDATE%29 http://www1.beuth-hochschule.de/FB_II/reports/Report-2019-004.pdf

Can we provide some pointers on README.md on how we can install AIF360 from the github source itself? Some ideas here: https://stackoverflow.com/questions/15268953/how-to-install-python-package-from-github

Include additional classification metrics

5

Include the following metrics: 1. Equalized odds difference: `max(|FPR_unpriv - FPR_priv|, |TPR_unpriv - TPR_priv|)` 2. Generalized equalized odds difference: `max(|GFPR_unpriv - GFPR_priv|, |GTPR_unpriv - GTPR_priv|)` 3. Generalized selection rate: mean...

Modify predict_proba in exponentiated_gradient_reduction.py

1

Per discussion in #215, predict_proba in exponentiated_gradient_reduction.py needs to be modified. See https://github.com/Trusted-AI/AIF360/pull/215#discussion_r523334486 for details.

The assert statement ```assert egr_aoe_race

Investigate if we can add support for folktables

2

folktables (https://github.com/zykls/folktables) is an "improved" version of UCI Adult dataset (which is derived from old US census data). It would be nice to natively incorporate this into the toolkit so...

Implement wasserstein fairness postprocessing

2

Implement algorithm 2 in http://proceedings.mlr.press/v115/jiang20a/jiang20a.pdf - a simple and efficient post-processing fairness method

See eqn 10.5 in http://trustworthymachinelearning.com/trustworthymachinelearning-10.htm First priority is to do this for sklearn compatible version and then for "classic" version.