alan-turing-institute / SBO_Pyro

A repository for prototyping structured bayesian optimisation approach with Pyro
MIT License
4 stars 1 forks source link

Meeting notes #1

Open tomaslaz opened 5 years ago

tomaslaz commented 5 years ago

26/09/2019

Current sources of variation between BOAT and pyro-boat:

  1. different acquisition function (LCB vs EI)
  2. different inference (SMC/MCMC vs SVI)
  3. different initial points for seeding the optimization
  4. different local optimization packages for optimizing surrogate
  5. different model (BOAT uses two different GPs for the separate residuals)

Performance differences:

This somewhat suggests that the GP is not the bottleneck, and instead something else is causing the delay over time with BOAT. Maybe this is due to inference, maybe due to the way the surrogate is optimized.