Draft
Conversation
Contributor
|
Turing.jl documentation for PR #2789 is available at: |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2789 +/- ##
===========================================
- Coverage 86.34% 20.21% -66.13%
===========================================
Files 22 23 +1
Lines 1435 1588 +153
===========================================
- Hits 1239 321 -918
- Misses 196 1267 +1071 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Member
Author
|
@penelopeysm I am running into the following issue when touching up my implementation. Suppose we run the following simple example using Random
using Turing
@model function linear_regression(x, y)
β ~ Normal(0, 1)
σ ~ truncated(Cauchy(0, 3); lower=0)
for t in eachindex(x)
y[t] ~ Normal(β * x[t], σ)
end
end
# condition the model
rng = MersenneTwister(1234)
x, y = rand(rng, 10), rand(rng, 10)
reg_model = linear_regression(x, y)
rng = MersenneTwister(1234)
particles = sample(rng, reg_model, SMC(0.5), 512, ensemble=MCMCSerial());Where I get the following error: ExceptionStackNot entirely sure why this is considering when I run this outside of the Turing environment (same exact code from |
Member
|
First, I found that |
penelopeysm
reviewed
Mar 18, 2026
I fixed this locally, but forgot to commit to the PR Co-authored-by: Penelope Yong <penelopeysm@gmail.com>
Member
Author
penelopeysm
reviewed
Mar 18, 2026
Member
|
This PR TuringLang/Libtask.jl#219 should fix the Libtask errors; with that PR + once you add the DynamicPPL imports, it runs correctly for me. If I had to take a guess, it was probably working for you before this because the variables weren't truly global (were they in a function, or some local scope?) |
Co-authored-by: Penelope Yong <penelopeysm@gmail.com>
Member
Author
All it took was the imports, though I appreciate the second look at Libtask. I don't think there's any global manipulation here, unless you count the task local storage trickery used by Libtask. |
Member
|
I don't understand that at all, but if there's no error now, I guess I'll take the win. |
Member
|
(The point about the globals was that if you run the code snippet above in the REPL, then x and y are global variables.) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Following the latest major release of DynamicPPL, I finally scrapped together my ideas for the long awaited restructuring of SMC within Turing.
Major Changes
TracedModelworks quite differently under the hood. While Libtask is still a central component of this mechanic, theproducecall from accumulation is delayed to ensureVarInfois caught up in terms of log-likelihoods.SMCno longer uses AbstractMCMC to interface; howeverParticleGibbsstill does.AbstractMCMCEnsemblenow interacts with SMC samplers along the reweighting step, which is designed to operate in parallel.TracedRNGhas been removed in favor of a taped global RNG manipulation to facilitate allocation efficient replayability of referenced trajectories in Particle Gibbs.TODO List
The original interface is for the most part replicated in its entirety. There are a handful of minor tweaks I need to make in order to better reinterface with some internal methods.
PartialLogDensity, or something to facilitate a partially observed modelsmcsamplewith MCMCChains/FlexiChains so that its consistent with the rest of the moduleNotes
This is all based on my self contained reimplementation here which serves more as a workspace for demonstration and experimentation. You can think of this PR as a subset of my personal repo; with that being said, I have a couple questions on proper integration:
Lastly, I would really appreciate help with the realization of particle rejuvenation. I have a demo over on TuringSMC which showcases a proof of concept.