Releases: TuringLang/SliceSampling.jl
Releases · TuringLang/SliceSampling.jl
v0.7.0
SliceSampling v0.7.0
Breaking changes
- The default value for
max_proposals
has been tightened to10_000
. This means that applying one of the samplers inSliceSampling
on a degenerate model might throw an error with a higher probability starting from v0.7.
Merged pull requests:
- Tweak default
max_proposals
limit (#16) (@Red-Portal) - Setup formatter (#17) (@Red-Portal)
v0.6.2
v0.6.1
SliceSampling v0.6.1
Merged pull requests:
- CompatHelper: bump compat for Turing in [weakdeps] to 0.34, (keep existing compat) (#11) (@github-actions[bot])
v0.6.0
v0.5.0
SliceSampling v0.5.0
- A default initializer has been added for Turing models. Thus,
sample
can be called without specifyinginitial_params
. - Univariate slice samplers can now be used directly on 1-dimensional targets through the
AbstractMCMC
interface.
v0.4.2
SliceSampling v0.4.2
v0.4.1
SliceSampling v0.4.1
v0.4.0
SliceSampling v0.4.0
Breaking Changes
- Univariate slice samplers now need to wrapped into
RandPermGibbs
to be used on multivariate targets.
New Features
- Added the hit-and-run sampling strategy for using univariate slice samplers on multivariate targets.
Merged pull requests:
- Add hit-and-run sampler, separate Gibbs sampling interface (#8) (@Red-Portal)
- enable docs for PRs (#9) (@Red-Portal)
v0.3.0
SliceSampling v0.3.0
- Added Gibbsian polar slice sampling
- Added optional limit on the maximum number of proposals to avoid infinite loops
Merged pull requests:
- Support Turing's
externalsampler
interface, returnTransition
instead of raw samples. (#1) (@Red-Portal) - Add window size check for univariate slice samplers (#2) (@Red-Portal)
- fix doc urls (#3) (@Red-Portal)
- Add Gibbsian polar slice sampler (#4) (@Red-Portal)
- fix README badges (#6) (@Red-Portal)
- Improve docs for GPSS (#7) (@Red-Portal)
Closed issues:
- TagBot trigger issue (#5)