Extreme Value Analysis
summary
Explains how extreme value analysis turns limited records (10–15 years) into one-in-50-year or even 500‑year return value (CI 2.50–5.82) estimates, shows the math behind the GEV/Gumbel family, compares block maxima vs peak‑over‑threshold pipelines, and demonstrates Python code using pi‑extremes (block sizing, fit by MLE/MOM, Monte Carlo for CIs) while noting Python tooling gaps versus R and listing advanced needs like covariate models and non‑stationary or multivariate implementations.
AI-written summary · about
chapters (18)
- 0:00Extreme Value Analysis intro
- 1:08Why EV analysis matters
- 2:35EV theory math basis
- 6:13Bulk data vs extremes
- 7:18Fisher-Tippett-Gnedenko theorem
- 8:43GEV distribution breakdown
- 11:54EV analysis in Python
- 14:23Extracting and visualizing extremes
- 15:51Return levels and confidence
- 17:17Model fit evaluation
- 20:11Fitting methods pitfalls
- 23:18Python EV packages
- 24:24EV analysis in R
- 26:11Advanced techniques gaps
- 26:53Coverage for spatial extremes
- 29:21Handling non-stationarity
- 30:03Machine learning and Extreme Value Analysis
- 30:48Q&A on EV applications
related talks
30mExtremes, outliers, and GOATS: on life in a lognormal world
29mModeling Extreme Events with PyMC
27mModelling the extreme using Quantile Regression
47mexpectation: A modern take on statistical A/B testing with e-values and martingales
34mPrediction intervals
01h 26mNew Trends in Estimation and Inference
08mProbabilistic Analysis of WCETs on RT Systems: When Your Worst Case Isn't Worst
19mA careful walk through probability distributions, using Python