Fix HMC in Gibbs issue with dynamic models #1217
Conversation
Codecov Report
@@ Coverage Diff @@
## master #1217 +/- ##
==========================================
+ Coverage 65.59% 65.91% +0.31%
==========================================
Files 25 25
Lines 1279 1282 +3
==========================================
+ Hits 839 845 +6
+ Misses 440 437 -3
Continue to review full report at Codecov.
|
|
Cool, thanks! |
| ∂logπ∂θ = gen_∂logπ∂θ(spl.state.vi, spl, model) | ||
| logπ = gen_logπ(spl.state.vi, spl, model) | ||
| spl.state.h = AHMC.Hamiltonian(metric, logπ, ∂logπ∂θ) | ||
| resize!(spl.state.z.θ, length(θ_old)) |
There was a problem hiding this comment.
It is worth noting that the vanilla HMC with identity mass matrix and pre-specified leapfrog step-size might be the only sensible HMC sampler for dynamic models for now, since the adaption schedules for step-size and mass matrix are not guaranteed to work well when the static dimensionality condition no longer holds.
There was a problem hiding this comment.
@trappmartin Maybe add a warning for this in the Infinite Mixture Model example?
There was a problem hiding this comment.
Good point. I'll open a PR for this.
* fix TuringLang#1095 * add test * fix bug * test PG and MH * fix tests * avoid duplicate line
This PR fixes #1095 by making sure the position in the HMC PhasePoint is of the right length and value.
CC: @xukai92 @trappmartin