Distinguishing Cause from Effect Using Quantiles: Bivariate Quantile Causal Discovery
Authors: Natasa Tagasovska, Valérie Chavez-Demoulin, Thibault Vatter
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To illustrate the effectiveness of our approach, we perform an extensive empirical comparison on both synthetic and real datasets. This study shows that b QCD is robust across different implementations of the method (i.e., the quantile regression), computationally efficient, and compares favorably to state-of-the-art methods. |
| Researcher Affiliation | Academia | 1HEC, University of Lausanne 2Swiss Data Science Center EPFL/ETHZ 3Statistics Department, Columbia University. |
| Pseudocode | No | The paper describes methods and processes but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code and datasets are available in the submitted supplementary package and https://github.com/tagas/b QCD. |
| Open Datasets | Yes | For simulated data, we first rely on the following scenarios (Mooij et al., 2016): SIM (without confounder)... There are 100 pairs of size n = 1000 in each of these datasets. For real data, we use the T ubingen CE benchmark (version Dec 2017), consisting of 108 pairs from 37 different domains, from which we consider only the 99 pairs that have univariate continuous or discrete cause and effect variables. |
| Dataset Splits | No | The paper mentions datasets used but does not explicitly provide specific training, validation, or test dataset splits, percentages, or methodology for data partitioning. |
| Hardware Specification | No | The paper discusses the computational efficiency and time taken for experiments (e.g., '7 minutes'), but it does not provide specific details about the hardware used (e.g., CPU/GPU models, memory, or number of cores). |
| Software Dependencies | Yes | R: A language and environment for statistical computing, 2017. URL https://www.r-project. org/. rvinecopulib: high performance algorithms for vine copula modeling, 2018. URL https://cran.r-project.org/ package=rvinecopulib. Eigen v3, 2010. Hmisc: Harrell Miscellaneous, 2017. URL https://cran.r-project.org/ package=Hmisc. |
| Experiment Setup | Yes | For estimating the overall quantile scores (aggregating multiple quantile levels), we use Legendre quadrature to approximate the integral over [0, 1]... Note that, to compute scores free of scale bias, the variables are transformed to the standard normal scale. The parameter m can be used to control for the trade-off between the computational complexity and the precision of the estimation. We recommend the value m = 3 which, makes it possible to capture variability in both location and scale. This was used to improve our results on real data, namely by setting the parameter m to 1 when n < 200 and m = 3 for the rest. |