Bayesian optimisation with scikit-learn
summary
Demonstrates Bayesian optimisation implemented with Gaussian processes as priors and an expected improvement acquisition function, using the scikit-learn GP module to iterate: fit GP, pick next theta by maximizing acquisition, evaluate, repeat. Claims it finds optima in fewer iterations than random or grid search but adds overhead and depends on kernel choice, parameter scaling and bounded search space; works well when single evaluations cost minutes–hours; warns random search can win when massively parallelised; points to Spearmint, hyperopt, SMAC and model-free alternatives like Hyperband.
AI-written summary · about
chapters (4)
related talks
28mBayesian Optimization- Fundamentals, Implementation, and Practice
24mUsing Bayesian Optimization to Tune Machine Learning Models
34mBayesian Optimization and it's application to Neural Networks
48mBayesian Optimization of Gaussian Processes with Applications to Performance Tuning
41mOptimizing Machine Learning Models
01h 30mA Primer on Gaussian Processes for Regression Analysis
01h 30mMaking Gaussian Processes Useful
26mScott Clark, CEO & Co-Founder, SigOpt at MLconf Seattle 2017