Releases: TuringLang/AdvancedHMC.jl
Releases · TuringLang/AdvancedHMC.jl
v0.2.1
v0.2.0
v0.2.0 (2019-07-23)
Closed issues:
- Get more comments and docstrings (#75)
- Add more informative error messages. (#71)
- HMC: Multinomial Sampling Instead of Slice Sampling (#62)
- Update gradient function to return both value and gradient (#56)
- Simplifying
Metric
types by usingAbstractPDMat
(#34) - RFC
find_good_eps
(#27) - DynamicHMC is (~5 times) faster than Turing's HMC implementation (#26)
Merged pull requests:
v0.1.9
v0.1.9 (2019-07-05)
Merged pull requests:
- Fix RFC typo (#78) (xukai92)
- Fixed bug preventing StanHMCAdaptor from adapting (#77) (chriselrod)
- Improving the performance and generality of adaptation (#74) (mohamed82008)
v0.1.8
v0.1.8 (2019-06-06)
Merged pull requests:
v0.1.7
v0.1.6
v0.1.5
Bugfix in `find_good_eps`.
v0.1.4
Improve trajectory (#58) * implement HMCDA and add abstract callback function * bugfix for H_new * remove comments * update version in Project.toml @JuliaRegistrator register()
Leapfrog bugfix and refactoring
- A typo in leapfrog integration was caught by @yebai!
- Some refactoring of leapfrog codes for future improvement
Improved numerical stability and enhanced NUTS interface
NaN
s are correctly handled now.- Hyper parameters for
NUTS
now can be set via the constructor. - Codes for the
AbstractMetric
type is simplified and improved.