Causal Discovery for Time Series

Framework for evaluating temporal observation-based causal discovery techniques on real and synthetic data.

Methods

• GrangerPW: https://www.jair.org/index.php/jair/article/view/13428
• GrangerMV: https://pubmed.ncbi.nlm.nih.gov/20481753
• TCDF: https://www.mdpi.com/2504-4990/1/1/19
• NAVARMLP: https://link.springer.com/chapter/10.1007/978-3-030-88942-5_35
• PCMCIParCorr: http://proceedings.mlr.press/v124/runge20a.html
• tsFCI: https://www.academia.edu/download/30837559/10.1.1.173.254.pdf#page=131
• CDNOD: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5617646
• VarLiNGAM: https://www.jmlr.org/papers/v11/hyvarinen10a.html
• TiMINo: https://proceedings.neurips.cc/paper/2013/hash/47d1e990583c9c67424d369f3414728e-Abstract.html
• DYNOTEARS: http://proceedings.mlr.press/v108/pamfil20a.html

Scripts

Test AD

Evaluates a given method against a batch of autonomous driving CSV data files. Assumes the CSV data files describe a two-agent convoy scenario.

usage: test_ad.py [-h] [--processor-count PROCESSOR_COUNT] [--verbose] [--max-time-lag MAX_TIME_LAG] [--sig-level SIG_LEVEL] method dataset variable

Parameters:

• method: Method to evaluate.
• dataset: Dataset folder under the "data" directory to acquire CSV data files from.
• variable: Sub-dataset to use under the dataset directory. Determines the variable of interest for each agent captured in the CSV data files.
• -h: Displays the help message for the script.
• --processor-count: Specifies the maximum number of processors to use. Only usable with methods which do not rely upon using an intermediary output file due to race condition risks.
• --verbose: Dictates that output should be verbose.
• --max-time-lag: Specifies the maximum time lag parameter for the selected method.
• --sig-level: Specifies the significance alpha parameter for the selected method.

Run All AD Tests

Evaluates all methods, across all datasets and sub-datasets for a given set of parameters. Primarily useful for setting off batch jobs.

usage: [SKIP_TO_METHOD=?] [OVERRIDE_METHODS=?] [OVERRIDE_DATASETS=?] [OVERRIDE_VARIABLES=?] ./run_all_ad_tests.sh

Parameters:

• SKIP_TO_METHOD: Any methods listed before the specified method in the script will be skipped.
• OVERRIDE_METHODS: Completely overrides the methods to evaluate listed in the script.
• OVERRIDE_DATASETS: Completely overrides the datasets to evaluate upon listed in the script.
• OVERRIDE_VARIABLES: Completely overrides the variables/sub-datasets to evaluate upon listed in the script.