Bayesian Non-parametric Models for Data Science using PyMC3
summary
Introduces Bayesian nonparametrics with working PyMC3 code and real datasets: Gaussian processes for regression and classification, a latent-GP Poisson model for coal disasters, a binomial GP for measles confirmation, spatial kriging on geostat data, and a 27k-runner Boston Marathon fit using a sparse approximation. Shows hyperparameters (rho, eta), uses optimization and NUTS MCMC with posterior predictive draws, reports runtimes (Boston ~1.5 minutes), explains cubic scaling and when approximations degrade fidelity; Dirichlet processes mentioned in the abstract but not developed.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:21BAYESIAN NON-PARAMETRIC MODELS FOR DATA SCIENCE USING PYMC3
- 3:22MOTIVATION
- 5:06STATISTICAL ASSUMPTIONS
- 6:34BAYESIAN INFERENCE
- 6:53BAYES FORMULA
- 8:12GAUSSIAN (NORMAL) DISTRIBUTION
- 12:22QUADRATIC
- 17:35SALMON RECRUITMENT MODEL
- 19:26MAKING PREDICTIONS WITH GAUSSIAN PROCESSES
- 21:52COAL MINING DISASTERS
- 24:44POSTERIOR HYPERPARAMETER ESTIMATES
- 27:30SCALING GAUSSIAN PROCESSES
- 27:49POSTERIOR COVARIANCE
- 28:57MARATHON TIMES 2015 Boston Marathon (26.598 finishers)
- 30:53MULTIDIMENSIONAL GP
- 31:56SPATIAL SAMPLES
- 33:36GAUSSIAN PROCESSES FOR DATA SCIENCE
- 35:12THE PYMC3 TEAM
related talks
44mAn introduction to PyMC3
01h 30mA Primer on Gaussian Processes for Regression Analysis
01h 27mProbabilistic Python: An Introduction to Bayesian Modeling with PyMC
39mProbablistic Programming Data Science with PyMC3
01h 33mMastering Gaussian Processes with PyMC
01h 30mAn introduction to Markov Chain Monte Carlo using PyMC3
42mSolving Real-World Business Problems with Bayesian Modeling
01h 31mProbabilistic Programming and Bayesian Inference in Python