PyMC v3.3 Release Notes
Release Date: 2018-01-09 // over 6 years ago-
๐ New features
- Improve NUTS initialization
advi+adapt_diag_grad
and addjitter+adapt_diag_grad
(#2643) - โ Added
MatrixNormal
class for representing vectors of multivariate normal variables - Implemented
HalfStudentT
distribution - ๐ New benchmark suite added (see http://pandas.pydata.org/speed/pymc/)
- ๐ Generalized random seed types
- โก๏ธ Update loo, new improved algorithm (#2730)
- ๐ New CSG (Constant Stochastic Gradient) approximate posterior sampling algorithm (#2544)
- ๐ Michael Osthege added support for population-samplers and implemented differential evolution metropolis (
DEMetropolis
). For models with correlated dimensions that can not use gradient-based samplers, theDEMetropolis
sampler can give higher effective sampling rates. (also see PR#2735) - ๐ Forestplot supports multiple traces (#2736)
- โ Add new plot, densityplot (#2741)
- ๐ DIC and BPIC calculations have been deprecated
- ๐จ Refactor HMC and implemented new warning system (#2677, #2808)
๐ Fixes
- ๐ Fixed
compareplot
to useloo
output. - ๐ Improved
posteriorplot
to scale fonts sample_ppc_w
now broadcastsdf_summary
function renamed tosummary
- โ Add test for
model.logp_array
andmodel.bijection
(#2724) - Fixed
sample_ppc
andsample_ppc_w
to iterate all chains(#2633, #2748) - โ Add Bayesian R2 score (for GLMs)
stats.r2_score
(#2696) and test (#2729). - SMC works with transformed variables (#2755)
- Speedup OPVI (#2759)
- ๐ Multiple minor fixes and improvements in the docs (#2775, #2786, #2787, #2789, #2790, #2794, #2799, #2809)
๐ Deprecations
- ๐ Old (
minibatch-
)advi
is removed (#2781)
- Improve NUTS initialization