v0.14.0

Turing v0.14.0

Diff since v0.13.0

Closed issues:

• Add LKJ Correlation Distribution (#924)
• Update docs for MAP/MLE estimation (#1293)
• Error using predict (#1294)
• Multivariate normal issues (#1295)
• MLE/MAP with start values gives ambiguity (#1298)
• Tutorial "Bayesian Hidden Markov Models" was not reproducible (#1302)
• Typo in Infinite Mixture Models tutorial (number of clusters formula) (#1304)
• 7 Errors during test Turing with Julia 1.4.2 (#1305)
• Make Gibbs components extensible (somewhat RFC...) (#1306)
• Incorrect handling of type parameter in HMC (#1308)
• Linear regression tutorial was not reproducible (#1313)
• no method matching reconstruct with MH on a vector of Dirichlet vectors (#1317)
• Documentation build CI is failing (#1319)
• Are there ways to break symmetries in Turing models? (#1328)
• type Wishart has no field c0 error (#1336)
• Unable to run "Using DynamicHMC" example (#1337)
• Problem with Quick Start example (#1339)
• Zygote compatibility (#1347)
• Progress bar doesn't work for NUTS sampler (#1350)
• Wrong prediction results on multivariate params (#1352)
• Extracting sampled parameters is broken (#1353)
• Tests failing due to the ambiguity error for \ with PDMat and TrackedArray (#1358)
• Truncated or truncated not supported under Zygote backend (#1361)
• Apply AD gradient if optimizer is a first-order one (#1365)
• Precompile Turing fails with AbstractChains not defined (#1366)
• Can not load Turing.jl, LoadError: UndefVarError: AbstractChains not defined (#1367)
• How to do non-centered parameterization in Turing? (#1372)
• log: DomainError with -960.1841700397322 (#1374)
• Option to discard burn-in for HMC/MH (#1379)
• Update docs (#1390)

Merged pull requests:

• Update MLE/MAP and parallel sampling docs (#1296) (@cpfiffer)
• Affine invariant ensemble sampler (#1297) (@cpfiffer)
• Allow users to provide a starting point for MLE/MAP (#1299) (@cpfiffer)
• Make GibbsComponent a simple trait (fix #1306) (#1307) (@phipsgabler)
• Fix #1308 (#1309) (@devmotion)
• Make Bijectors.jl lower-bound 0.7 instead of 0.6.4 (#1310) (@torfjelde)
• fix typo (#1316) (@ArnoStrouwen)
• Fix MH heuristics for vectors of multivariate distributions (#1318) (@devmotion)
• Added a stickbreak function in RandomMeasures. (#1320) (@luiarthur)
• Fix argument name (#1323) (@harryscholes)
• Fixes for MCMCChains 4 (#1324) (@devmotion)
• Update DistributionsAD and Distributions compatibilities (#1327) (@devmotion)
• Add CompatHelper secret (#1330) (@devmotion)
• Use and test DynamicPPL.alg_str (#1338) (@devmotion)
• Remove DEBUG (#1342) (@devmotion)
• Turing sampler docs (#1344) (@denainjs)
• residual sampling weird weighting (#1345) (@francescoalemanno)
• CompatHelper: bump compat for "EllipticalSliceSampling" to "0.3" (#1349) (@github-actions[bot])
• Fix for issue #1352 (#1357) (@torfjelde)
• Use AdvancedVI.jl for variational inference (#1362) (@torfjelde)
• Update guide.md (#1363) (@ludgerpaehler)
• Typos in advanced.md (#1364) (@ludgerpaehler)
• Add .jekyll-cache to gitignore (#1368) (@cpfiffer)
• Provide AD gradient for MLE/MAP (#1369) (@cpfiffer)
• Update performancetips.md (#1370) (@wupeifan)
• Add MH docstring (#1371) (@cpfiffer)
• Add some missing RNGs (#1373) (@devmotion)
• Add "Turing is fast" to splash page (#1375) (@cpfiffer)
• Simplify Turing.Variational and fix test error (#1377) (@devmotion)
• Require Julia 1.3 (#1378) (@devmotion)
• Fix argument type of a NUTS kwarg in docs (#1386) (@jangevaare)
• Use separate test environment (#1387) (@devmotion)
• CompatHelper: bump compat for "Optim" to "0.22" for package test (#1388) (@github-actions[bot])
• Documentation updates (#1391) (@devmotion)
• Update DynamicPPL (#1392) (@devmotion)