Testing Probabilistic Circuits

Authors: Yash Pralhad Pote, Kuldeep S Meel

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the practical efficiency of our algorithmic framework via a detailed experimental evaluation of a prototype implementation against a set of 475 PC benchmarks.
Researcher Affiliation Academia Yash Pote r Kuldeep S. Meel School of Computing, National University of Singapore
Pseudocode Yes The pseudocode of Teq is given in Algorithm 1. Algorithm 1 Teq(ϕ1, w1, ϕ2, w2, ε, η, δ)
Open Source Code Yes The accompanying tool, available open source, can be found at https://github.com/meelgroup/teq.
Open Datasets Yes We conducted experiments on a range of publicly available benchmarks arising from sampling and counting tasks4. Our dataset contained 100 d-DNNF circuits with weights. Footnote 4: https://zenodo.org/record/3793090
Dataset Splits No The paper describes experiments to evaluate the Teq algorithm on various benchmarks, not the training and validation of a machine learning model, hence standard train/validation/test splits are not applicable or described.
Hardware Specification Yes Our experiments were conducted on a high performance compute cluster with Intel Xeon(R) E5-2690 v3@2.60GHz CPU cores.
Software Dependencies No The paper states that the prototype was implemented in Python and uses WAPS3 [17], but does not provide specific version numbers for these software components.
Experiment Setup Yes We set the closeness parameter ε, farness parameter η and confidence δ for Teq to be 0.01, 0.2 and 0.01, respectively. The chosen parameters imply that if the input pair of probabilistic circuits are 0.01 close in d T V , then Teq returns Accept with probability atleast 0.99, otherwise if the circuits are 0.2 far in d T V , the algorithm returns Reject with probability at least 0.99. The number of samples required for Teq (indicated by the variable m as on line 2 of Algorithm 1) depends only on ε, η, δ and for the values we have chosen, we find that we require m = 294 samples.