keli95566 / explore_tigramite Goto Github PK

This project explores the possiblity of tigramite with different real-world dataset. It is in preparation of a user study that aim for improving the visualization of time series causal graph of tigramite.

├── src
│   ├── model     <- code of building prediction models.
│   ├── data      <- code for post-processing and pre-processing of raws dataset.
│   └── visualize <- source of visualization code.
│ 
├── data
│   ├── raw       <-raw data in the format of csv, etc.
│   │               (If an API to the source is avaliable commit the link.)
│   └── processed  <- processed data in the format of npz 
│                (The npz file can directly be put into tigramite dataframe)
│ 
├── model         <- built models based on tigramite for faster loading.
└── notebook      <- a collection of notebooks documenting experiments on different datasets. 

Note:

Keep a clean format of the notebook: ind-initial-name. Example: (01-kl-covid19-epi-exploration)

• notebook structure:
  • title and experiment descriptions.
  • data processing (loading data from Google drive, convert data to numpy format, etc).
  • causality analysis using Tigramite.
  • user study notes or verbal analysis.

• Conceptual Introduction

  • Causal Inference and Causal Discovery in Climate Science by Marlene Kretschmer, on Mar 2021: This talk gives a basic introduction to the difference of causal inference, causal discovery, and how to do correlation test, and independence test.
  • Causal inference in Earth system sciences by Jakob Runge, on April 2021: This talk gives a comprehensive overview of causal inference and an introduction to time series causal discovery.
  • Introducing Environmental Data Science: a new Open Access journal Launch event at NeurIPS 2020, how are AI and data science enhancing our understanding of the environment, including climate change?

• Algorithm Introduction:

  • Causal discovery in time series with unobserved confounders by Andreas Gerhardus (DLR Jena), on April 8, 2021: This talk gives a shorter overview of causal inference, and an introduction to Latent PCMCI algorithm.

  • Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series by Jakob Runge, on Oct 2020: This talk gives a brief introduction to the PCMIC+ Algorithm.