Causal Inference in Time Series via Supervised Learning

Authors: Yoichi Chikahara, Akinori Fujino

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In experiments, our method sufficiently outperformed the model-based Granger causality methods and the supervised learning method for i.i.d. data by using the same feature representation and the same classifier.
Researcher Affiliation Industry Yoichi Chikahara and Akinori Fujino NTT Communication Science Laboratories, Kyoto 619-0237, Japan chikahara.yoichi@lab.ntt.co.jp, fujino.akinori@lab.ntt.co.jp
Pseudocode No No pseudocode or algorithm blocks are present in the paper.
Open Source Code No The paper provides links to the code for existing methods (RCC, GCVAR, GCGAM, GCKER, TE) that are used for comparison, but not for the authors' proposed method (SIGC).
Open Datasets Yes The first test dataset was composed of five pairs of bivariate time series downloaded from the Cause-Effect Pairs database [Jakob, ], whose true causal relationships are reported in [Jakob, ] as X Y for three pairs and as X Y for the others. For instance, the River Runoff is a bivariate time series concerning average precipitation X and average river runoff Y , and the true causal relationship is regarded as X Y ." and "[Jakob, ] Zscheischler Jakob. Database with cause-effect pairs. https://webdav.tuebingen.mpg.de/cause-effect/.\" and \"We used the Saccharomyces cerevisiae (yeast) cell cycle gene expression dataset collected by [Spellman et al., 1998].
Dataset Splits Yes The number of trees is selected from {100, 200, 500, 1000, 2000} via 5-fold cross validation.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper mentions using a random forest classifier and Gaussian kernels, but does not provide specific version numbers for these software components or any other libraries used.
Experiment Setup Yes The number of trees is selected from {100, 200, 500, 1000, 2000} via 5-fold cross validation." and "For our method, we selected W = 12.\" and \"we set the number of features m = 100\" and \"Finally, we scaled each time series with mean 0 and variance 1.