This is the code for the paper "On Concept-Based Explanations in Deep Neural Networks" https://arxiv.org/abs/1910.07969".
- Please download all data from "gs://concept_discovery/data/" or "https://console.cloud.google.com/storage/browser/concept_discovery", and change the directory paths when loading data.
- To run the toy example, just run python3 toy_main.py.
- To create the toy example, just run python3 create_toy.py.
- To run the AwA example, just run python3 awa_main.py.
ipca.py is a general helper function for calculating the completeness of a given model, and toy_helper.py and awa_helper.py contain helper functions that are specific to the datasets.