gtx 1.0

A Ground Truth eXplanation Dataset

Ground Truth eXplanation Dataset

1. Please run the setup.sh file to downlad the datasets.
2. Please run conda env create --name hddds --file environment.yml to install the necessary packages.
3. Please find the notebook example.ipynb for the sample access to our dataset and potential use of dataloaders.
4. Please find the code for experiments in hddds.py (which is going to be slow). The detailed code is consist of three parts, and separated in three files:

• experiments/train.py: training the predictive models;
• experiments/get_explanations.py: generating the explanations using different feature atrribution methods;
• experiments/summary.py: evaluating the performance of the explanations against ground truth.

PS: If you ever wonder why there are so many 'd's in hddds, it is pinyin for Hao Da De DataSet (very large dataset) :)