Pointpriors (#663)

* implement pointwise_logpriors

* implement varwise_logpriors

* remove pointwise_logpriors

* revert dot_assume to not explicitly resolve components of sum

* docstring varwise_logpriores

use loop for prior in example

Unfortunately cannot make it a jldoctest, because relies on Turing for sampling

* integrate pointwise_loglikelihoods and varwise_logpriors by pointwise_densities

* record single prior components

by forwarding dot_tilde_assume to tilde_assume

* forward dot_tilde_assume to tilde_assume   for Multivariate

* avoid recording prior components on leaf-prior-context

and avoid recording likelihoods when invoked with leaf-Likelihood context

* undeprecate pointwise_loglikelihoods and implement pointwise_prior_logdensities

mostly taken from #669

* drop vi instead of re-compute vi

bgctw first forwared dot_tilde_assume to get a correct vi
and then recomputed it for recording component prior densities.

Replaced this by the Hack of torfjelde that completely drops vi and recombines the value, so that assume is called only once for each varName,

* include docstrings of pointwise_logdensities
pointwise_prior_logdensities
int api.md docu

* Update src/pointwise_logdensities.jl remove commented code

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Update src/pointwise_logdensities.jl remove commented code

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Update test/pointwise_logdensities.jl rename m to model

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Update test/pointwise_logdensities.jl remove unused code

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Update test/pointwise_logdensities.jl rename m to model

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Update test/pointwise_logdensities.jl rename m to model

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Update src/test_utils.jl remove old code

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* rename m to model

* JuliaFormatter

* Update test/runtests.jl remove interactive code

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* remove demo_dot_assume_matrix_dot_observe_matrix2 testcase

testing  higher dimensions better left for other  PR

* ignore local interactive development code

* ignore temporary directory holding local interactive development code

* Apply suggestions from code review: clean up comments and Imports

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* Apply suggestions from code review: change test of applying to chains on already used model

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>

* fix test on names in likelihood components

to work with literal models

* try to fix testset pointwise_logdensities chain

* Update test/pointwise_logdensities.jl

* Update .gitignore

* Formtating

* Fixed tests

* Updated docs for `pointwise_logdensities` + made it a doctest not
dependent on Turing.jl

* Bump patch version

* Remove blank line from `@model` in doctest to see if that fixes the
parsing issues

* Added doctest filter to handle the `;;]` at the end of lines for matrices

---------

Co-authored-by: Tor Erlend Fjelde <tor.github@gmail.com>
Co-authored-by: Tor Erlend Fjelde <tor.erlend95@gmail.com>

Author: bgctw

Project.toml

@@ -1,6 +1,6 @@
 name = "DynamicPPL"
 uuid = "366bfd00-2699-11ea-058f-f148b4cae6d8"
-version = "0.29.1"
+version = "0.29.2"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"

docs/src/api.md

@@ -124,10 +124,14 @@ Return values of the model function for a collection of samples can be obtained
 generated_quantities
 ```
 
-For a chain of samples, one can compute the pointwise log-likelihoods of each observed random variable with [`pointwise_loglikelihoods`](@ref).
+For a chain of samples, one can compute the pointwise log-likelihoods of each observed random variable with [`pointwise_loglikelihoods`](@ref). Similarly, the log-densities of the priors using
+[`pointwise_prior_logdensities`](@ref) or both, i.e. all variables, using
+[`pointwise_logdensities`](@ref).
 
 ```@docs
+pointwise_logdensities
 pointwise_loglikelihoods
+pointwise_prior_logdensities
 ```
 
 For converting a chain into a format that can more easily be fed into a `Model` again, for example using `condition`, you can use [`value_iterator_from_chain`](@ref).

src/DynamicPPL.jl

@@ -115,6 +115,8 @@ export AbstractVarInfo,
     # Convenience functions
     logprior,
     logjoint,
+    pointwise_prior_logdensities,
+    pointwise_logdensities,
     pointwise_loglikelihoods,
     condition,
     decondition,
@@ -181,7 +183,7 @@ include("varinfo.jl")
 include("simple_varinfo.jl")
 include("context_implementations.jl")
 include("compiler.jl")
-include("loglikelihoods.jl")
+include("pointwise_logdensities.jl")
 include("submodel_macro.jl")
 include("test_utils.jl")
 include("transforming.jl")