Releases: TuringLang/Turing.jl
Releases · TuringLang/Turing.jl
v0.30.1
Turing v0.30.1
Merged pull requests:
- Update HISTORY.md (#2149) (@sunxd3)
- Fix AbstractMCMC 5 compatibility (#2153) (@devmotion)
Closed issues:
- Transpilation of pure WinBUGS code when reimplementing Prior and Posterior Prediction (#2148)
v0.30.0
What's Changed
- Replaced specifying a global AD backend with
ADTypes.jl
. Users should now specify the desiredADType
directly in sampler constructors, e.g.,HMC(0.1, 10; adtype=AutoForwardDiff(; chunksize))
, orHMC(0.1, 10; adtype=AutoReverseDiff(false))
(false
indicates not to use compiled tape). - Interface functions such as
ADBackend
,setadbackend
,setadsafe
,setchunksize
, andsetrdcache
are deprecated and will be removed in a future release. - Removed the outdated
verifygrad
function. - Updated to a newer version of
LogDensityProblemsAD
(v1.7).
New Contributors
Full Changelog: v0.29.3...v0.30.0
v0.29.3
v0.29.2
Turing v0.29.2
Merged pull requests:
- CompatHelper: bump compat for SciMLBase to 2, (keep existing compat) (#2087) (@github-actions[bot])
- Fix + test for compiled ReverseDiff without linking (#2097) (@torfjelde)
- Fix for CI (#2098) (@torfjelde)
Closed issues:
- Dimensionality error when using MAP as initial point for sampling (#2079)
- Model using LKJCholesky fails with large covariance matrices (#2081)
- MCMCDistributed() not working properly - only worker 2 is sampling (#2082)
- Very high initial values for standard deviation (#2085)
- DynamicNUTS (via DynamicHMC) no longer works in Julia v1.9. (#2090)
LKJCholesky
does not work with compiled ReverseDiff.jl (#2091)- Compute (Bayesian) R2 for Turing models (#2093)
v0.29.1
Turing v0.29.1
Merged pull requests:
- Attach
varname_to_symbol
mapping toChains
(#2078) (@torfjelde)
v0.29.0
Turing v0.29.0
Closed issues:
- How to conduct Bayesian inference using an existing posterior samples as new prior distribution, as well as new observations in Turing? (#2069)
Merged pull requests:
- Reorganised some files. (#2055) (@yebai)
- Remove some obsolete utility exports. (#2070) (@yebai)
- Remove usage of
DynamicPPL.tonamedtuple
(#2071) (@torfjelde) FlattenIterator
is replaced byvarname_and_value_leaves
(#2072) (@yebai)- rename a few folders for clarity (#2073) (@yebai)
- Remove redundant docs link (#2075) (@willtebbutt)
Tracker
imports no longer needed. (#2076) (@yebai)
v0.28.3
Turing v0.28.3
Closed issues:
- method overwritten when I add Turing.jl pkg (#2061)
- Can't apply Bijectors.ordered to TDist(), says it's constrained (#2064)
- Incorrect bijector dimensions for models using SimplexBijector (#2065)
Merged pull requests:
- Fix incorrect output dimensions in stacked bijectors (#2066) (@bgroenks96)
v0.28.2
Turing v0.28.2
Closed issues:
- Benchmarking Turing against Stan on very simple models (#1283)
- Deadlinks (#1647)
- Turing slows down as the number of generic types increases (#1940)
- Turing checkpointing by callbacks (#1995)
- Allow LogDensityProblems.logdensity to take a named tuple (#2038)
- Saving generated values (#2059)
Merged pull requests:
- New feature: muliple iterations within Gibbs (#1881) (@xukai92)
- CompatHelper: bump compat for Bijectors to 0.13, (keep existing compat) (#2018) (@github-actions[bot])
- MH Constructor (#2037) (@JaimeRZP)
- CompatHelper: bump compat for NamedArrays to 0.10, (keep existing compat) (#2062) (@github-actions[bot])
- CompatHelper: bump compat for NamedArrays to 0.10 for package test, (keep existing compat) (#2063) (@github-actions[bot])