v0.6.18

v0.6.18 (2019-07-05)

Diff since v0.6.17

Closed issues:

• Domain Error on Develpment Branch (#814)
• Documentation Search always(?) leads to 404 (#806)
• Extracting docs from a closed PR (#805)
• Documentation: Code error in Guide (#795)
• Structured Stochastic Variational Inference (#794)
• "sample from the prior" vs "sample from generative model' (#786)
• AD error in logsumexp (#781)
• Testing the TuringBenchBot (#779)
• Disable numerical tests on Appveyor CI (#776)
• RFC Sampler type (#771)
• Move SGLD and SGHMC to AdvancedHMC.jl (#769)
• Improve code for initialising Sampler (#768)
• Make DynamicNUTS robust to numerical errors (#767)
• Turing v0.6.17 fails (#766)
• Release v0.6.17 (#764)
• Moving SGHMC and SGLD samplers to AdvancedHMC. (#748)
• Run Windows tests on Travis CI (#725)
• [RFC] Refactoring of the Sampler APIs (#723)
• Extract symbol and indices from varname string (#721)
• Add example illustrating how to run multiple chains. (#698)
• Scalable MH (#675)
• NUTS adapatation samples appear in full chain (#635)
• VarInfo performance and model types (#620)
• Variational Bayes inference? (#606)
• Introduce MCMCSampler type (#602)
• Win32 tests are very slow on AppVeyor (#530)
• Add additional keyword arguments to the sample function (#482)
• Debugging tools (#379)
• Refactoring VarInfo (#377)
• Remove Flat distribution (#315)
• Improving the speed of single program run (#294)
• Adapative MH (#287)

Merged pull requests:

• Update Turing to 0.6.18 (#838) (cpfiffer)
• Delete homepage-updater.jl (#835) (yebai)
• Update NUTS docstring (#832) (xukai92)
• Move benchmarks to this repo (#824) (KDr2)
• Set up CI with Azure Pipelines (#823) (yebai)
• Run benchmark on Travis-CI (#816) (KDr2)
• Minor update for #786 (#811) (cpfiffer)
• add a function to split a var string to symbol and inds (#809) (mohamed82008)
• Maybe fix the 404 error in the search bar (#808) (cpfiffer)
• Move AD tests to core testing suite (#807) (cpfiffer)
• Make AllUtils.jl inclusion copy-paste-friendly (#804) (mohamed82008)
• TS 3: Hook sample to TypedVarInfo (#803) (mohamed82008)
• Fix AHMC version before #791 is merged (#801) (xukai92)
• Improvements in CRP code (#800) (trappmartin)
• Hacks to make MvNormal work with Tracker (#798) (willtebbutt)
• Guide updates (#797) (cpfiffer)
• Change HMCDA default adaptation (#791) (xukai92)
• Pass verbose argument to AHMC (#790) (xukai92)
• Created stdlib folder - no functionality change. (#787) (yebai)
• Interface Implementation Part 1 (#785) (cpfiffer)
• Discard adaptation samples (#784) (cpfiffer)
• logsumexp AD workaround (#782) (mohamed82008)
• Disallow numerical tests on Appveyor (#778) (cpfiffer)
• Merge PMCMC samplers into one file - no functionality change. (#773) (yebai)
• Fix some bug in eval\_num from #740 (#770) (mohamed82008)
• TS 2: introduce TypedVarInfo and fix spl.info[:cache_updated] (#742) (mohamed82008)
• Replace HMC related codes using AHMC (#739) (xukai92)

v0.6.17

• Fixes for MvNormal distributions (#757)
• Modifications to the Project.toml fle (#763, #765)

EPS bug fix

• Fixes a bug in find_good_eps which was effecting HMC initialization (#744)
• Fixes a message about missing dependencies (#765)

Bugfixes and internal improvements.

This release fixes

• DynamicHMC interface #730
• CmdStan interface #729

and substantially improved unit and integration tests for Turing (#731)

BNP priors for random partitions

The long-awaited support for BNP priors for random partitions has finally arrived (#591)! This release introduces Dirichlet processes, Chinese restaurant processes, Pitman-Yor processes, and much more! Tutorials and documentation to follow.

Bugfixes

• MCMCChains indexing fixes #715
• Bug-fixing in dual averaging for HMC #709

Changed MCMCChain to MCMCChains

MCMCChain → MCMChains Changeover (#695)

Julia 1.1 Support

• Turing now supports Julia 1.1!
• Fixed a NegativeBinomial gradient issue (#680)
• Updated handling of the Turing.Chain object, support for missing values (#682)
• Upgrades Libtask to v0.2.5, required for Julia 1.1 compatibility (#685)
• Fixed a PMMH issue, matrix support test, and an array shape problem (#686)

BinomialLogit Bugfix

• Fixes a bug where VecBinomialLogit and BinomialLogit were not exported. See #676 for more info.

Bug Fixes and Easier Generative Models

• Fixes a CmdStan overflow issue in TuringBenchmarks (#662)
• Improved testing coverage for distributions (#621)
• Added an upper bound on step size for HMC-based samplers (#659, #671)
• Fixed an index parsing bug in the Chain type (#658)
• Fixed IPMCMC sampler bug (#663)
• Fixed automatic differentiation through nbinomlogpdf (#664)
• Improved the API on the backend for HMC samplers (#671)
• It's easier to draw from a prior now, so treating an existing Turing model as a generative one no longer requires a long list of default values in the model call (#644, #651). By passing in a Vector{Missing} object, the sampler will draw samples from the prior rather than the posterior. As an example:

# Import packages.
using Turing

# Define a simple Normal model with unknown mean and variance.
@model gdemo(x) = begin
    s ~ InverseGamma(2,3)
    m ~ Normal(0, sqrt(s))
    for i in eachindex(x)
        x[i] ~ Normal(m, sqrt(s))
    end
end

# Treat x as a vector of missing values.
model = gdemo(fill(missing, 2))
c = sample(model, HMC(500, 0.01, 5))