This repository holds the code for the NeurIPS 2022 paper Semantic Probabilistic Layers by Kareem Ahmed, Stefano Teso, Kai-Wei Chang, Guy Van den Broeck and Antonio Vergari
We introduce Semantic Probabilistic Layers, a drop-in replacement for the Softmax layer that guarantees the consistency of the neural network's predictions with a given set of symbolic constraints, while retaining probabilistic semantics, supporting arbitrary constraints, all while being tractable.
conda env create -f environment.yml
and if you encounter a pypsdd related error, running the following should solve the issue
pip install -vvv --upgrade --force-reinstall --no-binary :all: --no-deps pysdd
Each of the four tasks includes a .sh script for training and testing.
To use the CUB mini dataset, enter the --dataset hyperparameter as cub_mini. For full CUB, use cub_others. For other datasets, type the name followed by the dataset type, e.g. expr_FUN or enron_others.
