Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Extend gradlogpdf to MixtureModels #1827

Draft
wants to merge 15 commits into
base: master
Choose a base branch
from
Draft

Extend gradlogpdf to MixtureModels #1827

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
94cf6c0
gradlogpdf for mixtures
rmsrosa Jan 20, 2024
e7dd8ef
make them work on array of multiple samples
rmsrosa Jan 20, 2024
3930983
use iszero instead of > 0
rmsrosa Jan 20, 2024
65e3210
tests with a zero prior
rmsrosa Jan 20, 2024
af7dde6
update docs
rmsrosa Jan 20, 2024
6ed9476
no docstring
rmsrosa Jan 21, 2024
fe2f778
remove implicit broadcasting of gradlogpdf
rmsrosa Jan 23, 2024
56860a1
make gradlogpdf type stable... hopefully
rmsrosa Jan 23, 2024
b50f5eb
make gradlogpdf type stable
rmsrosa Jan 26, 2024
1d3c281
use iterate
rmsrosa Jan 27, 2024
7ccc7dc
update docstring for gradlogpdf
rmsrosa Jan 27, 2024
2c42066
increase coverage
rmsrosa Jan 27, 2024
c87d948
remove unnecessary test
rmsrosa Jan 27, 2024
53e0977
Apply suggestions from code review
rmsrosa Jan 29, 2024
4d4af7e
add tests for single distribution "mixtures"
rmsrosa Jan 29, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
55 changes: 47 additions & 8 deletions src/mixtures/mixturemodel.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -367,15 +367,31 @@ function _mixlogpdf!(r::AbstractArray, d::AbstractMixtureModel, x)
return r
end

function _mixgradlogpdf1(d::AbstractMixtureModel, x)
pdfdx = pdf(d, x)
glp = !iszero(pdfdx) ? sum(pi * pdf(d.components[i], x) .* gradlogpdf(d.components[i], x) for (i, pi) in enumerate(probs(d)) if (!iszero(pi) && !iszero(pdf(d.components[i], x)))) / pdfdx : pdfdx
return glp
end

pdf(d::UnivariateMixture, x::Real) = _mixpdf1(d, x)
logpdf(d::UnivariateMixture, x::Real) = _mixlogpdf1(d, x)
gradlogpdf(d::UnivariateMixture, x::Real) = _mixgradlogpdf1(d, x)

function gradlogpdf(d::UnivariateMixture, x::Real)
cp = components(d)
pr = probs(d)
pdfx1 = pdf(cp[1], x)
pdfx = pr[1] * pdfx1
_glp = pdfx * gradlogpdf(cp[1], x)
glp = (!iszero(pr[1])) && (!iszero(pdfx)) ? _glp : zero(_glp)
@inbounds for i in Iterators.drop(eachindex(pr, cp), 1)
rmsrosa marked this conversation as resolved.
Show resolved Hide resolved
if !iszero(pr[i])
pdfxi = pdf(cp[i], x)
if !iszero(pdfxi)
pipdfxi = pr[i] * pdfxi
pdfx += pipdfxi
glp += pipdfxi * gradlogpdf(cp[i], x)
end
end
end
if !iszero(pdfx) # else glp is already zero
glp /= pdfx
end
Comment on lines +398 to +400
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is it correct to return a gradlogpdf of zero if x is not in the support of the mixture distribution?

Copy link
Author

@rmsrosa rmsrosa Jan 29, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wondered about that, but decided to follow the current behavior already implemented. For example:

julia> insupport(Beta(0.5, 0.5), -1)
false

julia> logpdf(Beta(0.5, 0.5), -1)
-Inf

julia> gradlogpdf(Beta(0.5, 0.5), -1)
0.0

I don't know. If it is constant -Inf, then the derivative is zero (except that (-Inf) - (-Inf) is not defined, but what matters is that the rate of change is zero...)

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Most implementations return zero outside the support:
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/beta.jl#L126
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/chi.jl#L99
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/chisq.jl#L94
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/exponential.jl#L97

and so on (Frechet, Gamma, ...), but some (Arcsin and Kumaraswamy) seem to just extend what happens at the edges of the support:
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/arcsine.jl#L97
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/kumaraswamy.jl#L91

while LogitNormal seems to be the only one that yields NaN:
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/laplace.jl#L101

Meanwhile, Laplace seems to be the only one to error at a singular point:
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/laplace.jl#L101

while Uniform, for example, is zero everywhere including the singular points.
https://github.com/JuliaStats/Distributions.jl/blob/master/src/univariate/continuous/uniform.jl#L86

I think giving an error is a bit drastic, but yielding either zero or NaN (outside the support or at the singular points) seems reasonable to me.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe we should make this a separate issue and hopefully standardize the behavior.

return glp
end

_pdf!(r::AbstractArray{<:Real}, d::UnivariateMixture{Discrete}, x::UnitRange) = _mixpdf!(r, d, x)
_pdf!(r::AbstractArray{<:Real}, d::UnivariateMixture, x::AbstractArray{<:Real}) = _mixpdf!(r, d, x)
Expand All @@ -386,7 +402,30 @@ _logpdf(d::MultivariateMixture, x::AbstractVector{<:Real}) = _mixlogpdf1(d, x)
_pdf!(r::AbstractArray{<:Real}, d::MultivariateMixture, x::AbstractMatrix{<:Real}) = _mixpdf!(r, d, x)
_logpdf!(r::AbstractArray{<:Real}, d::MultivariateMixture, x::AbstractMatrix{<:Real}) = _mixlogpdf!(r, d, x)

gradlogpdf(d::MultivariateMixture, x::AbstractVector{<:Real}) = _mixgradlogpdf1(d, x)
function gradlogpdf(d::MultivariateMixture, x::AbstractVector{<:Real})
cp = components(d)
pr = probs(d)
pdfx1 = pdf(cp[1], x)
pdfx = pr[1] * pdfx1
glp = pdfx * gradlogpdf(cp[1], x)
if ( iszero(pr[1]) || iszero(pdfx) )
glp .= zero(eltype(glp))
rmsrosa marked this conversation as resolved.
Show resolved Hide resolved
end
@inbounds for i in Iterators.drop(eachindex(pr, cp), 1)
if !iszero(pr[i])
pdfxi = pdf(cp[i], x)
if !iszero(pdfxi)
pipdfxi = pr[i] * pdfxi
pdfx += pipdfxi
glp .+= pipdfxi * gradlogpdf(cp[i], x)
end
end
end
if !iszero(pdfx) # else glp is already zero
glp ./= pdfx
end
return glp
end

## component-wise pdf and logpdf

Expand Down
10 changes: 6 additions & 4 deletions test/gradlogpdf.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,22 +28,24 @@ using Test

# Test for gradlogpdf on univariate mixture distributions against centered finite-difference on logpdf

x = [-0.2, 0.3, 0.8, 1.3, 10.5]
delta = 0.001
x = [-0.2, 0.3, 0.8, 1.0, 1.3, 10.5]
delta = 0.0001

for d in (
MixtureModel([Normal(-4.5, 2.0)], [1.0]),
MixtureModel([Exponential(2.0)], [1.0]),
MixtureModel([Uniform(-1.0, 1.0)], [1.0]),
MixtureModel([Normal(1//1, 2//1), Beta(2//1, 3//1), Exponential(3//2)], [3//10, 4//10, 3//10]),
MixtureModel([Normal(-2.0, 3.5), Normal(-4.5, 2.0)], [0.0, 1.0]),
MixtureModel([Beta(1.5, 3.0), Chi(5.0), Chisq(7.0)], [0.4, 0.3, 0.3]),
MixtureModel([Exponential(2.0), Gamma(9.0, 0.5), Gumbel(3.5, 1.0), Laplace(7.0)], [0.3, 0.2, 0.4, 0.1]),
MixtureModel([Logistic(-6.0), LogNormal(5.5), TDist(8.0), Weibull(2.0)], [0.3, 0.2, 0.4, 0.1])
)
xs = filter(s -> insupport(d, s), x)
xs = filter(s -> all(insupport.(d, [s - delta, s, s + delta])), x)
glp1 = gradlogpdf.(d, xs)
glp2 = ( logpdf.(d, xs .+ delta) - logpdf.(d, xs .- delta) ) ./ 2delta
@test isapprox(glp1, glp2, atol = delta)
@info "Testing `gradlogpdf` on $d"
@test isapprox(glp1, glp2, atol = 0.01)
end

# Test for gradlogpdf on multivariate mixture distributions against centered finite-difference on logpdf
Expand Down
Loading