On Causal Discovery in the Presence of Deterministic Relations
Authors: Loka Li, Haoyue Dai, Hanin Al Ghothani, Biwei Huang, Jiji Zhang, Shahar Harel, Isaac Bentwich, Guangyi Chen, Kun Zhang
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted extensive experiments on both simulated and real-world datasets to show the efficacy of our proposed method. |
| Researcher Affiliation | Collaboration | Loka Li1 , Haoyue Dai2 , Hanin Al Ghothani1, Biwei Huang3, Jiji Zhang4, Shahar Harel5, Isaac Bentwich5, Guangyi Chen1,2, Kun Zhang1,2 1 Mohamed bin Zayed University of Artificial Intelligence 2 Carnegie Mellon University, 3 University of California San Diego 4 The Chinese University of Hong Kong, 5 Quris AI |
| Pseudocode | Yes | Algorithm 1 DGES: Determinism-aware Greedy Equivalent Search |
| Open Source Code | Yes | The code is available at https://github.com/lokali/DGES.git. |
| Open Datasets | Yes | The true DAGs are simulated using the Erdös Rényi model [33] with the number of edges equal to the number of variables. One is the pharmacokinetics dataset [34], which is an open database for pharmacokinetics information from clinical trials. The other one is the US census Public Use Microdata Sample (PUMS). We follow the data preprocessing procedure outlined in [35], which is a modern version of the UCI Adult data set [36]. |
| Dataset Splits | Yes | There are two types of likelihoods as introduced in the paper, for computational efficiency, we choose the generalized score with cross-validated (CV) likelihood. SGS(Vi, PAG i ) = 1 q=1 ℓ(F (q) i |D(q) 0,i ). |
| Hardware Specification | Yes | We run our method and the other baseline methods in Ubuntu 20.04 LTS 64-bit System with Intel(R) Xeon(R) Silver 4214 2.20GHz 64 CPU.s |
| Software Dependencies | No | We implement this method based on the Causal-learn package https://github.com/py-why/causal-learn [62]. Our implementation is based on the code from https://github.com/juangamella/ges. It mentions 'Ubuntu 20.04 LTS 64-bit System' but lacks specific version numbers for other key software components (e.g., Python, libraries). |
| Experiment Setup | Yes | During the first phase, when we aim to detect the DC and Min DCs and check whether a variable can be deterministically represented by some others, we set that if the term Σu 2 HS < 1e 3, although theoretically the value should exactly be zero. Meanwhile, the regularization parameter for the kernel ridge regression is set to 1e 10. The second phase of our method is to run modified GES, and the setting is by default. The penalty parameter for controlling the sparsity is set to 1. |