[Diff since v0.15.20](TuringLang/Turing.jl@v0.15.20...v0.15.21)

**Closed issues:**
- Multivariate Distributions & Parameter Inference (TuringLang#1570)
- Example from tutorial does not work (TuringLang#1608)

**Merged pull requests:**
- Use correct statistical notation in Hello World (TuringLang#1605) (@palday)
- replace s with s² for variances in docs (TuringLang#1606) (@palday)
- CompatHelper: bump compat for "AdvancedMH" to "0.6" (TuringLang#1611) (@github-actions[bot])
- CompatHelper: bump compat for "AdvancedPS" to "0.2" for package test (TuringLang#1612) (@github-actions[bot])
- Fix AbstractMCMC deprecations (TuringLang#1614) (@devmotion)

[Diff since v0.15.19](TuringLang/Turing.jl@v0.15.19...v0.15.20)

**Merged pull requests:**
- Add timing information (TuringLang#1604) (@cpfiffer)

[Diff since v0.15.18](TuringLang/Turing.jl@v0.15.18...v0.15.19)

**Closed issues:**
- Predict errors with multiple chains/threads (TuringLang#1301)
- Is init_theta keyword working? (TuringLang#1588)
- The Infinite Mixture Models tutorial error (TuringLang#1589)
- Problems with MAP estimation using Optim (TuringLang#1594)
- ZygoteAD dot operator fails again (TuringLang#1595)
- Mention Zulip channel in the README? (TuringLang#1596)
- PG - not retaining reference particle? (TuringLang#1598)

**Merged pull requests:**
- Retain reference particle properly (TuringLang#1599) (@devmotion)

[Diff since v0.15.17](TuringLang/Turing.jl@v0.15.17...v0.15.18)

**Closed issues:**
- Odd issue with certain arraydist contents, ReverseDiff, and sum (TuringLang#1573)

**Merged pull requests:**
- Make SGHMC more flexible (TuringLang#1586) (@haozhangphd)

[Diff since v0.15.16](TuringLang/Turing.jl@v0.15.16...v0.15.17)

**Closed issues:**
- _params_to_array is very slow (TuringLang#1584)

**Merged pull requests:**
- Make `_params_to_array` faster, fixes TuringLang#1584 (TuringLang#1585) (@haozhangphd)

[Diff since v0.15.15](TuringLang/Turing.jl@v0.15.15...v0.15.16)

**Closed issues:**
- Wrong prediction results on multivariate params (TuringLang#1352)

**Merged pull requests:**
- Fix for TuringLang#1352 (TuringLang#1567) (@torfjelde)

[Diff since v0.15.14](TuringLang/Turing.jl@v0.15.14...v0.15.15)

**Merged pull requests:**
- Bump AbstractMCMC major version (TuringLang#1579) (@torfjelde)

[Diff since v0.15.13](TuringLang/Turing.jl@v0.15.13...v0.15.14)

**Merged pull requests:**
- Run tests on Julia 1.6 (TuringLang#1577) (@devmotion)
- Improve `maybe_link!` for MH sampler (TuringLang#1582) (@torfjelde)

[Diff since v0.15.12](TuringLang/Turing.jl@v0.15.12...v0.15.13)

**Closed issues:**
- LoadError: KeyError: key :name not found when using @model (TuringLang#1561)
- Issues with conditionals inside loops? (TuringLang#1562)

**Merged pull requests:**
- Update "Contributing" link (TuringLang#1564) (@tjburch)
- Run tests with Julia 1.5 instead of Julia 1.6 (TuringLang#1568) (@devmotion)
- CompatHelper: bump compat for "Bijectors" to "0.9" (TuringLang#1569) (@github-actions[bot])
- Update Documentation.yml (TuringLang#1572) (@devmotion)

[Diff since v0.15.11](TuringLang/Turing.jl@v0.15.11...v0.15.12)

**Closed issues:**
- Segmentation fault when running tests on Julia v1.6.0-rc1 (TuringLang#1554)
- RWMH inside Gibbs sampling (TuringLang#1556)

**Merged pull requests:**
- Update members.yml (TuringLang#1555) (@xukai92)
- Make MH proposal generation more sane (TuringLang#1557) (@cpfiffer)