Making Gaussian Processes Useful
summary
Tutorial turns Gaussian processes into a usable tool: hierarchical binomial and count models on baseball batting and slugging data, implemented in PyMC, use latent GPs for an age effect and an HSGP approximation to cut cubic cost to roughly n+M so sampling runs in minutes. Issues (divergences, identifiability between GP intercept and global intercept) resolve by switching to a Matern 5/2 kernel, informative length‑scale priors and a centered parameterization; sampling uses the nutpie sampler and yields a peak slugging around age 26.
AI-written summary · about
chapters (11)
- 0:00Gaussian processes tutorial
- 1:06Bayesian inference basics
- 13:51Baseball batting model
- 20:55What are GPs and why use them
- 33:19Hilbert space GPs for efficiency
- 40:32Transition to demo
- 41:15Batting model with GPs
- 51:16GPs as hierarchical extensions
- 1:03:06Debugging GP fitting
- 1:11:17Slugging vs age modeling
- 1:24:28Summary and Q&A
related talks
44mActually using GPs in practice with PyMC
01h 33mMastering Gaussian Processes with PyMC
01h 30mA Primer on Gaussian Processes for Regression Analysis
42mBayesian Non-parametric Models for Data Science using PyMC3
40mBdrygp: A Boundary-Integrated Gaussian Process Model For Computer Code Emulation
39mBayesian optimisation with scikit-learn
31mBut what is a Gaussian process? Regression while knowing how certain you are
01h 23mProbabilistic Graphical Models, HMMs using PGMPY