Minimum Cost Intervention Design for Causal Effect Identification

Authors: Sina Akbari, Jalal Etesami, Negar Kiyavash

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental For evaluation, we generated the causal graphs using the Erdos-Renyi generative model (Erd os & R enyi, 1960) as follows. For a given number of vertices n, we fixed a causal order over the vertices. Then, directed edges were sampled with probability p = 0.35 and bidirected edges were sampled with probability q = 0.25 between the vertices, mutually independently. The set S was selected randomly among the last 5% of the vertices in the causal order such that G[S] is a c-component. Intervention costs of vertices were chosen independently at random from {1, 2, 3, 4}. See Appendix F for further details of the evaluation setup. Our performance measures are runtime and normalized regret. Normalized regret of a given subset A is defined by (C(A) C )/C , where C denotes the optimal min-cost solution. The results are depicted in Figure 3. Each curve and its confidence interval is obtained by averaging over 40 trials.
Researcher Affiliation Academia Sina Akbari 1 Jalal Etesami 1 Negar Kiyavash 1 1College of Management of Technology, EPFL. Correspondence to: Sina Akbari <sina.akbari@epfl.ch>.
Pseudocode Yes Algorithm 1 Find Hhull(S, G), G[S] is a c-component.
Open Source Code Yes The implementations of all the algorithms proposed in this work can be found at https://github.com/Sina Akbarii/min_cost_intervention/tree/main.
Open Datasets Yes F.1. Benchmark Structures. In this section, we evaluate our algorithms on graphs corresponding to real-world problems, namely the Barley (Kristensen & Rasmussen, 1997), Water (Jensen et al., 1989) and Mehra (Vitolo et al., 2018) structures.
Dataset Splits No The paper describes the generation of synthetic graphs and the use of benchmark graph structures for evaluation, but it does not specify explicit train/validation/test dataset splits in the conventional sense for machine learning models.
Hardware Specification No The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not list any specific software dependencies with version numbers.
Experiment Setup Yes For evaluation, we generated the causal graphs using the Erdos-Renyi generative model (Erd os & R enyi, 1960) as follows. For a given number of vertices n, we fixed a causal order over the vertices. Then, directed edges were sampled with probability p = 0.35 and bidirected edges were sampled with probability q = 0.25 between the vertices, mutually independently. The set S was selected randomly among the last 5% of the vertices in the causal order such that G[S] is a c-component. Intervention costs of vertices were chosen independently at random from {1, 2, 3, 4}. See Appendix F for further details of the evaluation setup.