Efficient sampling from mixture models#547

Closed

kadeng wants to merge 14 commits intopymc-devs:masterfrom

Commits on Jun 5, 2014

Changed bound() function so that it efficiently deals with scalars and vector inputs without calculating unneccessary and possibly invalid expressions
Kai Londenberg
committed
Added full Metropolis Hastings step method for arbitrary (asymmetric) proposal distributions. Allows to create efficient custom samplers.
Kai Londenberg
committed
Added Gaussian Mixture distribution, which can be efficiently sampled from using Metropolis Hastings and a custom Proposal Distribution
Kai Londenberg
committed
Renamed Metropolis Hastings example
Kai Londenberg
committed
Fixed problem when sampling from transformed Dirichlet with hyperprior containing 1's.
Kai Londenberg
committed
Added an example which uses a GMM as a prior for another distribution. Fixed some problems that occurred.
Kai Londenberg
committed

Commits on Jun 6, 2014

Reverted back to old (switch) implementation of bound, due to problems with other examples
Kai Londenberg
committed
Reverted back to old (switch) implementation of bound, due to problems with other examples
Kai Londenberg
committed
Fixed case when arguments to bound_scalar have differing dtype
Kai Londenberg
committed
Fixed case when arguments to bound_scalar have differing dtype
Kai Londenberg
committed
Small changes in GMM example
Kai Londenberg
committed

Commits on Jun 14, 2014

Fixed bug in Wihart distribution. Added Inverse Wishart distribution. Added references to papers describing how to use these for both.
Kai Londenberg
committed
Added constructor for a special noninformative prior for covariance matrices
Kai Londenberg
committed
Small changes to the InverseWishart and the special covariance prior
Kai Londenberg
committed