search

rmculpepper / gamble

Probabilistic Programming
BSD 2-Clause "Simplified" License
40 stars 10 forks source link

Latest batch of HMC changes. #7

Closed lambdageek closed 10 years ago

lambdageek commented 10 years ago
  1. Support wrapping (observe-at ...) in a label or derivative form for HMC.
  2. If HMC leapfrog step moves into a region with 0 probability, abort the proposal (reject and retry).
  3. Compute the HMC acceptance threshold precisely as in the R. Neal paper. Previous expression was wrong.
  4. Add an example of doing linear regression using HMC. It seems to converge quite well now.
rmculpepper commented 10 years ago

Merged (cherry-picked, actually)