Causal Bounds in Quasi-Markovian Graphs

Authors: Madhumitha Shridharan, Garud Iyengar

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our heuristic on queries for the quasi-markovian graphs in Figure 1g, 1h and 1j in Zaffalon et al. (2020a). ... Overall, we evaluate our heuristic on 9 causal inference problems. ... In Sections 5.2.2 and 5.2.1, we benchmark Algorithm 1 against the most recent alternatives proposed in literature to our knowledge: both variations of Approx LP and the branch-and-bound algorithm in Duarte et al. (2021).
Researcher Affiliation Academia 1Department of Industrial Engineering and Operations Research, Columbia University, New York, USA.
Pseudocode Yes Algorithm 1 Frank Wolfe Heuristic for Lower Bound
Open Source Code No The paper does not include an unambiguous statement about releasing source code for the described methodology or a direct link to a code repository.
Open Datasets No We evaluate our heuristic on queries for the quasi-markovian graphs in Figure 1g, 1h and 1j in Zaffalon et al. (2020a). ... For each problem, we randomly generate 50 values of q as our ground truth. Then for each value, we compute the corresponding true value of the query, and input data distribution P(VCk|VPCk ), k = 1, . . . , m. The paper references graph structures from a previous work but does not provide concrete access or citation details for the generated input data distributions used in their experiments.
Dataset Splits No The paper describes generating 'ground truth' values and running the heuristic multiple times (T=10 random restarts) but does not specify explicit training, validation, or test dataset splits in the conventional sense for model evaluation.
Hardware Specification No The paper mentions optimization solvers like Gurobi and SCIP, but does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies Yes using readily available optimization solvers like Gurobi (Gurobi Optimization, LLC, 2023).
Experiment Setup Yes The procedure is initialized by generating a random feasible solution q(0). ... The entire procedure is repeated T times, each with a random initialization of the first iterate q(0). ... We then run Algorithm 1 on the input data distribution with T = 10, and check if the output bounds contain the true query value.