Axiomatic effect propagation in structural causal models
Authors: Raghav Singal, George Michailidis
JMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We showcase the application of RSV on two challenging problems on causality (causal overdetermination and causal unfairness). 6.2 Numerical illustration of Algorithms 3 and 4 7. Applications |
| Researcher Affiliation | Academia | Raghav Singal EMAIL Operations and Management Science Tuck School of Business at Dartmouth College Hanover, NH 03755, USA George Michailidis EMAIL Department of Statistics and Data Science University of California, Los Angeles Los Angeles, CA 90095, USA |
| Pseudocode | Yes | We formally state our recursive definition of RSV in Algorithm 1 (and sub-routine Algorithm 2). |
| Open Source Code | No | The paper does not provide a direct link to source code or explicitly state that source code is being released for the methodology described. |
| Open Datasets | Yes | In particular, we focus on the DAG shown in Figure 21, which is motivated by the graph topologies discovered in recent works (Zhang et al., 2017; Wu et al., 2019), by employing the PC algorithm (Spirtes et al., 2000) on the adult dataset in the UCI repository (Dua and Graff, 2017). |
| Dataset Splits | No | The paper does not specify dataset splits (e.g., train/test/validation percentages or sample counts) for the experiments. The applications involve simulating outcomes based on structural equations rather than splitting a dataset for traditional machine learning tasks. |
| Hardware Specification | Yes | With 10,000 noise samples (used to evaluate expected outcome), it took around 10 minutes on a 3.8 GHz 8-core Intel i7 machine with 16 GB memory to compute RSV for all the edges. |
| Software Dependencies | No | The paper mentions the use of 'the PC algorithm' but does not specify any software names with version numbers for its implementation or other dependencies. |
| Experiment Setup | Yes | In particular, we focus on the DAG shown in Figure 21, which is motivated by the graph topologies discovered in recent works (Zhang et al., 2017; Wu et al., 2019), by employing the PC algorithm (Spirtes et al., 2000) on the adult dataset in the UCI repository (Dua and Graff, 2017). ( 4 log(age) + N(0, 32) if race = 0 5 log(age) + N(0, 32) if race = 1 (20a) To compute the expected outcome for a given instance of the source variables (race, age, and sex), we generate 10,000 samples of the noise, which we found to be large enough to provide a stable estimate. In Figure 23a, we show the RSV 11 corresponding to the SEM of (20) with a background (race(1), age(1), sex(1)) = (0, 40, 0) and a foreground (race(2), age(2), sex(2)) = (1, 40, 1) and unfairness parameters (a1, a3) = (1, 1). In Figure 24, we show the RSV for (a1, a3) {(0, 0), (0, 0.2), (0.2, 0), (0.2, 0.2)}. |