Fix gamma mixture #222
Fix gamma mixture #222
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
| @@ -13,7 +13,7 @@ convenient functions to generate samples and weights to approximate expectations | |||
| - `rng`: random number generator objects | |||
| - `nsamples`: number of samples generated by default | |||
| """ | |||
| struct ImportanceSamplingApproximation{T, R} | |||
| struct ImportanceSamplingApproximation{T, R} <: AbstractFormConstraint | |||
There was a problem hiding this comment.
The ImportanceSamplingApproximation should not be declared as an AbstractFormConstraint. This is wrong from the semantic point of view. There should be an extra MomentsApproximationFormConstraint (or something similar).
There was a problem hiding this comment.
|
|
||
| a = m^2 / v | ||
| b = m / v | ||
| return GammaShapeRate(a, b) |
There was a problem hiding this comment.
This should be changed somehow
|
|
||
| constrain_form(::ImportanceSamplingApproximation, distribution::Any) = distribution | ||
|
|
||
| make_form_constraint(::Type{ImportanceSamplingApproximation}, args...; kwargs...) = |
There was a problem hiding this comment.
This should be removed in favor of separate MomentsApproximationFormConstraint
| @@ -65,3 +65,7 @@ end | |||
|
|
|||
| Distributions.pdf(dist::GammaShapeRate, x::Real) = (rate(dist)^shape(dist)) / gamma(shape(dist)) * x^(shape(dist) - 1) * exp(-rate(dist) * x) | |||
| Distributions.logpdf(dist::GammaShapeRate, x::Real) = shape(dist) * log(rate(dist)) - loggamma(shape(dist)) + (shape(dist) - 1) * log(x) - rate(dist) * x | |||
|
|
|||
| function Random.rand(rng::AbstractRNG, dist::GammaShapeRate) | |||
There was a problem hiding this comment.
Lets also add more rand methods (which accepts size and inplace versions).
| @average_energy GammaShapeRate (q_out::Any, q_α::GammaDistributionsFamily, q_β::Any) = begin | ||
| mean(loggamma, q_α) - mean(q_α) * mean(log, q_β) - (mean(q_α) - 1.0) * mean(log, q_out) + mean(q_β) * mean(q_out) | ||
| end | ||
| @average_energy GammaShapeRate (q_out::Any, q_α::Union{PointMass, SampleList, GammaDistributionsFamily}, q_β::Any) = |
There was a problem hiding this comment.
I wonder if can simply put Any there
Initially, this PR aimed to solve #84.
It enables importance sampling for the
shapeparameters of the Gamma Mixture node (see supporting notebook onGamma Mixture).We have filled in the missing functions for
ImportanceSamplingApproximation.We have added the missing
randfunction for theGammaShapeRatetype.Average energy functions for
GammaShapeRatewere fused into a single function.