A practical guide to computing uncertainty estimates for data analysis, written from a hacker's perspective. Covers multiple methods including normal distribution confidence intervals, Beta distribution for binary outcomes, bootstrapping, maximum likelihood estimation, and Markov Chain Monte Carlo (MCMC) using the emcee library. Uses a concrete time series example throughout, with Python code using numpy, scipy, seaborn, and PyMC3. Distinguishes between data distribution, parameter uncertainty, and prediction uncertainty.
Table of contents
The hacker's guide to uncertainty estimatesLet’s get startedDistribution of the data vs uncertaintyConfidence intervals when all outcomes are 0 or 1RegressionBootstrapping, rebootedMarkov chain Monte Carlo methodsWrapping upFinallySort: