experiments

Repository containing experiments for evaluation of TrustyAI Explainability Toolkit

notebooks

Notebooks to reproduce available experiments:

• Original LIME implementation (discrete and continuous features setting) impact-score eval: https://colab.research.google.com/drive/1jLe-tdtE7uGQ0KIKMG4PWJjDgPPDn5W0#scrollTo=KjIRWtglSX0C

scripts

Train NAM model on FICO dataset

Install nam library from https://github.com/AmrMKayid/nam:

1. git clone https://github.com/AmrMKayid/nam
2. cd nam
3. pip install .

cd to experiments directory and run the training script

1. python train_nam_pytorch_fico.py

Inside saved_models/NAM_FICO you will find NAM checkpoints that can be used for inference.