Fix for #1352

[torfjelde]

Fixes #1352 and increases the lower-bound of DynamicPPL (the recent version includes bug-fixes for predict and generated_quantities).

[devmotion]

The tests fail since Libtask_jll does not support Julia 1.6.0 yet - it will be fixed by JuliaPackaging/Yggdrasil#2176 but the PR depends on JuliaPackaging/Yggdrasil#1987 (see #1554 for more details).

[devmotion]

Maybe we should run Turing tests with Julia 1.5, here and in DynamicPPL?

[torfjelde]

Maybe we should run Turing tests with Julia 1.5, here and in DynamicPPL?

Sounds like a good idea to me 👍

[torfjelde]

Just for reference: this is now only waiting on packages to get up-to-date with Bijectors.jl and running #1568

[devmotion]

The test error is strange (negative standard deviation in a Normal distribution), do you know what's going on there?

[torfjelde]

The test error is strange (negative standard deviation in a Normal distribution), do you know what's going on there?

Very strange indeed. And no, I don't. It shouldn't have anything to do with this particular PR though 😕

[torfjelde]

It's all the Gibbs-sampler btw

[torfjelde]

This should finally be good to go now, once tests pass 👍

[coveralls]

Pull Request Test Coverage Report for Build 739138834

• 6 of 6 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.02%) to 78.184%

---

Totals
Change from base Build 738567682:	0.02%
Covered Lines:	1111
Relevant Lines:	1421

---

💛 - Coveralls

[codecov]

Codecov Report

Merging #1567 (3504535) into master (be40a19) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1567      +/-   ##
==========================================
+ Coverage   78.16%   78.18%   +0.01%     
==========================================
  Files          23       23              
  Lines        1429     1421       -8     
==========================================
- Hits         1117     1111       -6     
+ Misses        312      310       -2     

Impacted Files	Coverage Δ
src/inference/Inference.jl	84.16% <100.00%> (-0.99%)	⬇️
src/inference/hmc.jl	76.47% <0.00%> (+1.17%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update be40a19...3504535. Read the comment docs.

[devmotion]

Looks good, just had a minor suggestion. It seems you also have to update the version number 🙂

[torfjelde]

What in the world. Gibbs is failing again

[torfjelde]

I'm very confused by that test. Nothing that I did in the fix for MH and linking changed the behavior of MH(), which is NOT linked and uses StaticProposal. But now it results in v2 in the test at https://github.com/TuringLang/Turing.jl/blob/master/test/inference/mh.jl#L149-L152 vanishing, which makes the test fail o.O

Is it okay if I just disable that particular comparison for now and make an issue for it? Really want to get the PR merged, and the only thing holding it back is tests unrelated to this PR failing due to random seed 😅

[yebai]

Is it okay if I just disable that particular comparison for now and make an issue for it? Really want to get the PR merged, and the only thing holding it back is tests unrelated to this PR failing due to random seed 😅

Happy to merge this PR as-is and open an issue for the failed test.