Generic support for uni gmm #615

xukai92 · 2017-05-27T14:41:34Z

Add generic support for uni-gmm

ararslan · 2017-05-27T21:11:26Z

These are now parametric types, you'll have to modify the constructors to account for the type parameter. The 0.5-compatible syntax is

function (::Type{UnivariateGMM{T}}){T}(...)
   ...
end

(Much nicer syntax for this is available in 0.6 but unfortunately it isn't compatible with 0.5.)

xukai92 · 2017-05-29T21:00:29Z

Thank you for pointing this. I will make the change soon. I have a problem which I don't know if I can ask here - actually I failed to run the tests for Distributions.jl on my local, with error shown below

Pkg.test("Distributions")
INFO: Computing test dependencies for Distributions...
ERROR: unsatisfiable package requirements detected: no feasible version could be found for package: ColorVectorSpace

Do you know why this happens?

ararslan · 2017-05-30T04:33:07Z

I've seen that error once before with a different package. If Pkg.update() shows unexpected warnings, you can always cd ~/.julia/v0.x/METADATA && git fetch origin && git reset --hard origin/metadata-v2 (where 0.x is your current Julia version, e.g. 0.6). If none of that works, I've heard that a solution that works is to delete the contents of ~/.julia/v0.x...

xukai92 · 2017-07-27T20:55:22Z

Hi Alex, I fix the constructor now.

I stoped working on this PR because I also meet some dependent problems - a group of functions used by uni gmm seems didn't support Dual before, and I didn't manage to fix all of them. But today after I fix the constructor and re-run the them, and the tests seems pass on my local now.

ararslan · 2017-07-27T21:26:46Z

src/mixtures/unigmm.jl

    prior::Categorical

-    function UnivariateGMM(ms::Vector{Float64}, ss::Vector{Float64}, pri::Categorical)
+    (::Type{UnivariateGMM{T}}){T}(ms::Vector{T}, ss::Vector{T}, pri::Categorical) = begin


Why change from the long-form function definition? That's generally preferable to begin blocks for style.

ararslan · 2017-07-27T21:27:03Z

src/mixtures/unigmm.jl

    end
 end

+UnivariateGMM{T<:Real}(ms::Vector{T}, ss::Vector{T}, pri::Categorical) = UnivariateGMM{T}(K, ms, ss, pri)


K isn't defined here

xukai92 · 2017-07-28T14:59:26Z

Thx Alex. I found that I messed up 3 Distributions.jl forks on my local so I actually run tests on another commit. I will give another try to fix it.

xukai92 · 2017-07-28T15:24:31Z

Ok, I think I meet the problem I see before: a lot of functions in mixturemodel.jl are not generic.

cdf related functions sees not generic support
- I would like to skip these first and do pdf related frist
I separated the test for generic unigmm and commented our cdf related functions now
- I tried to make some pdf related functions in mixturemodel.jl and now I face some numerical overflow?

julia test/mixture.jl
    testing UnivariateMixture
Test Failed
  Expression: logpdf(g,X[i]) ≈ mix_lp0[i]
   Evaluated: -Inf isapprox -2.0463780454938103
ERROR: LoadError: There was an error during testing
 in record(::Base.Test.FallbackTestSet, ::Base.Test.Fail) at test.jl:397
 in do_test(::Base.Test.Returned, ::Expr) at test.jl:281
 in test_mixture(::Distributions.MixtureModel{Distributions.Univariate,Distributions.Continuous,Distributions.Normal}, ::Int64, ::Int64) at mixture.jl:56
 in include_from_node1(::String) at loading.jl:488
 in include_from_node1(::String) at sys.dylib:?
 in process_options(::Base.JLOptions) at client.jl:265
 in _start() at client.jl:321
 in _start() at sys.dylib:?
while loading /Users/kai/projects/Turing/exps/Distributions.jl/test/mixture.jl, in expression starting on line 212

Can you give me some help and some general suggestions how to make things in mixturemodel.jl generics

matbesancon · 2018-09-19T21:49:44Z

Hi @xukai92 what's the status on this? Maybe you can keep going on this PR on Julia 1.0 and the latest Distributions master branch

matbesancon · 2019-02-27T12:45:04Z

ping @xukai92 to move on this :)

xukai92 · 2019-02-28T17:10:15Z

@matbesancon Thanks for pinging me! I'm on it now :)

xukai92 · 2019-02-28T19:45:06Z

I disabled the tests for generic support of GMM for rand(g, n) as this is not supported for Normal, i.e. below fails:

julia> d = Normal(Dual(0), Dual(1))
Normal{Dual{Nothing,Int64,0}}(μ=Dual{Nothing}(0), σ=Dual{Nothing}(1))

julia> rand(d, 10)
ERROR: MethodError: no method matching Float64(::Dual{Nothing,Float64,0})
Closest candidates are:
  Float64(::Real, ::RoundingMode) where T<:AbstractFloat at rounding.jl:194
  Float64(::T<:Number) where T<:Number at boot.jl:741
  Float64(::Int8) at float.jl:60
  ...
Stacktrace:
 [1] convert(::Type{Float64}, ::Dual{Nothing,Float64,0}) at ./number.jl:7
 [2] setindex!(::Array{Float64,1}, ::Dual{Nothing,Float64,0}, ::Int64) at ./array.jl:767
 [3] rand! at /Users/kai/projects/Distributions.jl/src/univariates.jl:167 [inlined]
 [4] rand at /Users/kai/projects/Distributions.jl/src/univariates.jl:160 [inlined]
 [5] rand(::Normal{Dual{Nothing,Int64,0}}, ::Int64) at /Users/kai/projects/Distributions.jl/src/genericrand.jl:24
 [6] top-level scope at none:0

xukai92 · 2019-02-28T19:50:57Z

@matbesancon Not sure why some tests for Hypergeometric fail.

matbesancon · 2019-03-01T09:14:29Z

yes sadly we need #836 merged for this first

matbesancon · 2019-03-05T14:55:23Z

not sure if we should do it here or in a following PR, but it would be good to replace Vector by a new implementation

xukai92 · 2019-03-06T03:00:45Z

Do yo mean making everything typed by AT, T where AT<:AbstractVector{T}, T<:Real?

matbesancon · 2019-03-06T14:25:14Z

yes exactly

matbesancon · 2019-03-06T14:26:09Z

Also what was done in #837 for multivariate gaussians

matbesancon · 2019-03-13T19:52:41Z

ping @xukai92 all good on this?

matbesancon · 2019-03-13T19:53:26Z

src/mixtures/mixturemodel.jl

@@ -135,13 +135,13 @@ function MixtureModel(components::Vector{C}, prior::Categorical) where C<:Distri
    MixtureModel{VF,VS,C}(components, prior)
 end

-MixtureModel(components::Vector{C}, p::Vector{Float64}) where {C<:Distribution} =
+MixtureModel(components::Vector{C}, p::Vector{T}) where {C<:Distribution,T<:Real} =


Should we allow for any <:AbstractVector{<:Real}?

Is this the only place we need to change? Or we actually need make AT<: AbstractVector part of the parameter of type MixtureModel?

yes exactly

src/mixtures/unigmm.jl

xukai92 · 2019-03-14T21:52:39Z

@matbesancon suggestions fixed! Please take another review.

matbesancon · 2019-03-15T09:05:30Z

Approved, merged in 12 hours if no objection

xukai92 · 2019-03-15T09:35:15Z

Thanks for the help @matbesancon !

mschauer

No objection.

matbesancon · 2019-03-17T16:25:19Z

woops, 12 hours became 2 days, merged. Thanks for the contribution @xukai92

xukai92 added 2 commits May 27, 2017 15:40

Make uni gmm support generic type

9caf4a9

Add generic test

ddf88e6

xukai92 mentioned this pull request May 27, 2017

Generic support for GMM type in Distributions.jl TuringLang/Turing.jl#266

Closed

xukai92 added 4 commits July 12, 2017 16:53

Merge branch 'master' into generic-support-for-uni-gmm

3b7d232

Disable generic test

85a41bf

Fix constructor

227efb8

Add generic test back

a825926

ararslan reviewed Jul 27, 2017

View reviewed changes

Fix K and try to make mixture.jl generic

0e4f5af

xukai92 added 4 commits February 28, 2019 17:19

Merge branch 'master' into generic-support-for-uni-gmm

9b572a7

Support Real

ebd7164

enable cdf tests

baf67d5

merge tests

90d0905

matbesancon reviewed Mar 13, 2019

View reviewed changes

matbesancon reviewed Mar 14, 2019

View reviewed changes

src/mixtures/unigmm.jl Outdated Show resolved Hide resolved

matbesancon reviewed Mar 14, 2019

View reviewed changes

src/mixtures/unigmm.jl Outdated Show resolved Hide resolved

matbesancon reviewed Mar 14, 2019

View reviewed changes

src/mixtures/unigmm.jl Outdated Show resolved Hide resolved

xukai92 added 2 commits March 14, 2019 21:47

allow mean and std to have differetn types

7f41b0f

allow p to be abstract vector

875b4e5

matbesancon requested a review from ararslan March 15, 2019 09:00

matbesancon approved these changes Mar 15, 2019

View reviewed changes

matbesancon requested a review from mschauer March 15, 2019 09:05

mschauer reviewed Mar 15, 2019

View reviewed changes

matbesancon merged commit 73e35a2 into JuliaStats:master Mar 17, 2019

simonbyrne mentioned this pull request Mar 19, 2019

Remove UnivariateGMM #844

Open

xukai92 mentioned this pull request Mar 20, 2019

AD fails for MixtureModel TuringLang/Turing.jl#722

Closed

xukai92 mentioned this pull request Mar 28, 2019

Make return type same as parameter instead of data #853

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Generic support for uni gmm #615

Generic support for uni gmm #615

xukai92 commented May 27, 2017

ararslan commented May 27, 2017

xukai92 commented May 29, 2017

ararslan commented May 30, 2017

xukai92 commented Jul 27, 2017

ararslan Jul 27, 2017

ararslan Jul 27, 2017

xukai92 commented Jul 28, 2017

xukai92 commented Jul 28, 2017

matbesancon commented Sep 19, 2018

matbesancon commented Feb 27, 2019

xukai92 commented Feb 28, 2019

xukai92 commented Feb 28, 2019

xukai92 commented Feb 28, 2019

matbesancon commented Mar 1, 2019

matbesancon commented Mar 5, 2019

xukai92 commented Mar 6, 2019

matbesancon commented Mar 6, 2019

matbesancon commented Mar 6, 2019

matbesancon commented Mar 13, 2019

matbesancon Mar 13, 2019

xukai92 Mar 13, 2019

matbesancon Mar 14, 2019

xukai92 commented Mar 14, 2019

matbesancon commented Mar 15, 2019

xukai92 commented Mar 15, 2019

mschauer left a comment

matbesancon commented Mar 17, 2019

Generic support for uni gmm #615

Generic support for uni gmm #615

Conversation

xukai92 commented May 27, 2017

ararslan commented May 27, 2017

xukai92 commented May 29, 2017

ararslan commented May 30, 2017

xukai92 commented Jul 27, 2017

ararslan Jul 27, 2017

Choose a reason for hiding this comment

ararslan Jul 27, 2017

Choose a reason for hiding this comment

xukai92 commented Jul 28, 2017

xukai92 commented Jul 28, 2017

matbesancon commented Sep 19, 2018

matbesancon commented Feb 27, 2019

xukai92 commented Feb 28, 2019

xukai92 commented Feb 28, 2019

xukai92 commented Feb 28, 2019

matbesancon commented Mar 1, 2019

matbesancon commented Mar 5, 2019

xukai92 commented Mar 6, 2019

matbesancon commented Mar 6, 2019

matbesancon commented Mar 6, 2019

matbesancon commented Mar 13, 2019

matbesancon Mar 13, 2019

Choose a reason for hiding this comment

xukai92 Mar 13, 2019

Choose a reason for hiding this comment

matbesancon Mar 14, 2019

Choose a reason for hiding this comment

xukai92 commented Mar 14, 2019

matbesancon commented Mar 15, 2019

xukai92 commented Mar 15, 2019

mschauer left a comment

Choose a reason for hiding this comment

matbesancon commented Mar 17, 2019