Optimizing NOTEARS Objectives via Topological Swaps

Authors: Chang Deng, Kevin Bello, Bryon Aragam, Pradeep Kumar Ravikumar

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in terms of achieving a better score. Additionally, our method can also be used as a post-processing algorithm to significantly improve the score of other algorithms. Code implementing the proposed method is available at https://github.com/duntrain/topo.
Researcher Affiliation Academia 1Booth School of Business, University of Chicago, USA. 2Machine Learning Department, Carnegie Mellon University, USA.
Pseudocode Yes Algorithm 1 TOPO; Algorithm 2 TOPO; Algorithm 3 FINDPARAMS; Algorithm 4 UPDATESORT
Open Source Code Yes Code implementing the proposed method is available at https://github.com/duntrain/topo.
Open Datasets No The paper states: "For each simulation, we generated n = 1000 samples for graphs with d {10; 20; 50; 100} nodes." and describes generating random graphs and sampling data based on models. This indicates synthetic data generation rather than the use of pre-existing, publicly available datasets with concrete access information.
Dataset Splits No The paper states it generated samples but does not specify dataset splits (e.g., percentages or counts) for training, validation, or testing, nor does it refer to predefined splits with citations.
Hardware Specification No The paper mentions: "We also thank the University of Chicago Research Computing Center for assistance with the calculations carried out in this work." However, it does not provide specific hardware details such as GPU/CPU models, memory, or specific cloud instances used for running the experiments.
Software Dependencies No The paper mentions software implementations like "FGS and PC are standard baseline...py-causal package", "NOTEARS...Python code", "KKTS...Python code", and "GOLEM...Python and Tensorflow code". However, it does not specify exact version numbers for any of these software components or libraries, which is required for reproducible description.
Experiment Setup Yes For TOPO, we use the least-square loss Q(W, X) = 1 2n X XW 2 F without any regularization for all noise type. We also use the polynomial acyclicity penalty h(A) = Tr((I + A/d)d) d (Yu et al., 2019) and h(A) = log det(I A) (Bello et al., 2022), because it is faster and more accurate than h(A) = Tr(e A) d (Zheng et al., 2018). For the choices of ssmall, slarge, s0, Table 7 summarizes the suggested hyerparameters.