milliondegree/pxai

PXAI

Before compile, add the path of the boost library to Makefile, e.g., -I /usr/local/include

Unzip mnist_train.csv at ./data/mnist/mnist_train.csv.zip

Run make all to compile three executable files in ./bin.

Run ./bin/cprov-maintain-query-overhead to estimate PXAI's provenance maintenance and query performance when applied to probabilistic graphical models (PGM).

Run ./bin/cprov-approx-ice-counter-overhead to estimate PXAI's XAI performance when applied to PGM.

Run ./bin/credit-score-mlp-test to estimate PXAI's performance when applied to multi-layer perceptron (MLP).

Run ./bin/mnist-kmeans-maintain-query-test to estimate PXAI's provenance maintenance performance when applied to k-Means.

Run ./bin/mnist-kmeans-test to estimate PXAI's ML deletion performance when applied to k-Mneas.

Run a test case of PGM: ./bin/pxai -o ./data/hypertext-class/sample7/sample71.obs -p ./data/hypertext-class/sample7/prov/sample71.txt -q topic_Department_29 -i all It computes the influences of the output to explain topic(Department, 29) through Individual Conditional Expectation (ICE).

Run make to rebuild the project and run a test case.