"Causal Discovery in Python"
summary
Compares concrete Python packages on the altitude/latitude/temperature example and reports what actually works: causal-learn runs fast with a one-line API and often recovers the altitude→temperature and latitude→temperature V-structure but shows seed-dependent flips; py-tetrad wraps Java Tetrad and scales to larger graphs and reliably finds the edges; cdt uses pairwise additive-noise methods, needs a complete graph input and misorients this example; causalnex gradient methods produce wrong directions; tigramite is only for time series.
AI-written summary · about
chapters (16)
- 0:00Introduction to causal discovery
- 1:33What is causal discovery
- 3:18When to use causal discovery
- 7:43Why causal discovery is hard
- 9:31Choosing the right algorithm
- 12:04Worked example: altitude, latitude, temperature
- 13:30Tetrad and CMU Python libraries
- 14:55Demo: causal-learn on example
- 16:42Demo: py-tetrad on example
- 18:28Causal Discovery Toolbox review
- 19:35Demo: CDT on example
- 20:44causalNex and gradient-based methods
- 21:31Demo: causalNex on example
- 22:17TigerJython and time-series focus
- 23:02Summary and tool recommendations
- 24:02Q&A: Predictions vs causal models
related talks
43mA Practical Guide to Causality in Python (For The Perplexed)
31mPython and the Holy Grail of Causal Inference
31mIntroduction to Causal Inference using pgmpy
06mCausal Machine Learning – What’s in it for Data Scientists?
01h 12mIntroduction to Causal Inference
29m🧠🧹 Causality - Mental Hygiene for Data Science
36mWhat-if- Causal reasoning meets Bayesian Inference
26mCausal inference with Causallib