Conversation
|
AdvancedHMC.jl documentation for PR #447 is available at: |
|
Tests pass locally, not sure what's going wrong with CI🤔 |
|
CI errors look like not our faults. I'll take a deeper look shortly, thanks! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #447 +/- ##
==========================================
+ Coverage 75.44% 76.18% +0.73%
==========================================
Files 21 21
Lines 1230 1285 +55
==========================================
+ Hits 928 979 +51
- Misses 302 306 +4 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
nsiccha
left a comment
nsiccha
left a comment
There was a problem hiding this comment.
Implementation seems to not handle preconditioning, data subsetting, or the LD part of SGLD properly.
| grad = last(logdensity_and_gradient(θ)) | ||
|
|
||
| stepsize = spl.stepsize(i) | ||
| θ .+= (stepsize / 2) .* grad .+ sqrt(stepsize) .* randn(rng, eltype(θ), length(θ)) |
There was a problem hiding this comment.
As before, I don't know the algorithm (or AdvancedHMC.jl), but to me it sounds as if the preconditioner has to play a role in this update?
There was a problem hiding this comment.
Also, it doesn't seem as if data subsetting is even considered?
| "Polynomial step size function." | ||
| stepsize::S | ||
| "Number of leapfrog steps." | ||
| n_leapfrog::Int |
There was a problem hiding this comment.
Shouldn't this always be 1?
3 participants
Fix: #60