New Distribution: MvDiscreteNonParametric #1424

davibarreira · 2021-11-16T01:01:51Z

I've created a new distribution which implements the DiscreteNonParametric distribution to the multivariate case.
The distribution followed the guidelines for new distributions (one can sample, calculate mean, var, cov, etc).
If this PR is accepted, I'll submit another PR with documentation.

Highlights:

mvdiscretenonparametric.jl - Has the implementation of the new MvDiscreteNonParametric distribution;
Dependency on ArraysOfArrays - MvDiscreteNonParametric uses ArraysOfArarys in order to properly work with vector of vectors, where each vector is a "sample" (point-mass).

The new distribution does not have a cdf yet. I'm working on a more efficient way to perform the computations instead of naively sorting and comparing everything.

…ssing.

devmotion · 2021-11-16T01:09:18Z

src/multivariate/mvdiscretenonparametric.jl

@@ -0,0 +1,153 @@
+struct MvDiscreteNonParametric{T <: Real,P <: Real,Ts <:  ArrayOfSimilarArrays{T},Ps <: AbstractVector{P}} <: DiscreteMultivariateDistribution


Can we make this more general and change it to something like

Suggested change

struct MvDiscreteNonParametric{T <: Real,P <: Real,Ts <: ArrayOfSimilarArrays{T},Ps <: AbstractVector{P}} <: DiscreteMultivariateDistribution

struct GeneralDiscreteNonParametric{VF,T,P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} <: Distribution{VF,Discrete}

? Then it could also be used instead of DiscreteNonParametric for the univariate case and we could define the alias

const DiscreteNonParametric{T<:Real,P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} = GeneralDiscreteNonParametric{Univariate,T,P,Ts,Ps}

to make it a non-breaking change.

And one could also define the alias

Suggested change

struct MvDiscreteNonParametric{T <: Real,P <: Real,Ts <: ArrayOfSimilarArrays{T},Ps <: AbstractVector{P}} <: DiscreteMultivariateDistribution

const MvDiscreteNonParametric{T<:AbstractVector{<:Real},P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} = GeneralDiscreteNonParametric{Multivariate,T,P,Ts,Ps}

Can you expand on how the GeneralDiscreteNonParametric shoudl work? It's not clear to me how this VF work, for example, I tried doing the following:

struct GeneralDiscreteNonParametric{VF,T,P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} <: Distribution{VF,Discrete} support::Ts p::Ps function GeneralDiscreteNonParametric{VF,T,P,Ts,Ps}(support::Ts, p::Ps) where {T,P <: Real,Ts <: AbstractVector{<:AbstractVector{T}},Ps <: AbstractVector{P}} length(support) == length(p) || error("length of `support` and `p` must be equal") isprobvec(p) || error("`p` must be a probability vector") allunique(support) || error("`support` must contain only unique value") new{VF,T,P,Ts,Ps}(support, p) end end

But of course I get an error, since VF is not defined.

Also, should the GeneralDiscreteNonParametric dispatch to MvDiscreteNonParametric and DiscreteNonParametric? Would something like CategoricalDistributions.jl be inside GeneralDiscreteNonParametric?

@devmotion , should I leave the functions for MvDiscreteNonParametric or should I drop it and just do everything inside the GeneralMvDiscreteNonParametric?

But of course I get an error, since VF is not defined.

You have to provide it, like

struct GeneralDiscreteNonParametric{VF,T,P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} <: Distribution{VF,Discrete} support::Ts p::Ps function GeneralDiscreteNonParametric{VF,T,P,Ts,Ps}(support::Ts, p::Ps; check_args=true) where {VF,T,P<: Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} if check_args length(support) == length(p) || error("length of `support` and `p` must be equal") isprobvec(p) || error("`p` must be a probability vector") allunique(support) || error("`support` must contain only unique values") end new{VF,T,P,Ts,Ps}(support, p) end end

Then you could define a univariate discrete non-parametric distribution with GeneralDiscreteNonParametric{Univariate,Float64,Float64,Vector{Float64},Vector{Float64}}([-10.0, -3.4, 2.1, 3.5], [0.4, 0.2, 0.1, 0.1]). But, of course, users shouldn't have to do this. Therefore we could use the aliases (also for the multivariate case) and define outer constructors such as

function DiscreteNonParametric(support::AbstractVector{<:Real}, p::AbstractVector{<:Real}; issorted=false) _p = issorted ? p : sort(p) return DiscreteNonParametric{eltype(support),eltype(_p),typeof(support),typeof(_p)}(support, _p) end

Besides Multivariate and Univariate, what kind of distribution you intend to keep under GeneralDiscreteNonParametric? Categorical Distributions?

They are already a type alias of DiscreteNonParametric, so if DiscreteNonParametric works they should just work automatically.

src/multivariate/mvdiscretenonparametric.jl

devmotion · 2021-11-16T01:12:17Z

src/multivariates.jl

+    for i in Base.OneTo(size(m, 2))
+        _rand!(rng, smp, view(m, :, i))


Can you revert the unrelated changes?

devmotion · 2021-11-16T01:14:45Z

Project.toml

@@ -4,6 +4,7 @@ authors = ["JuliaStats"]
 version = "0.25.29"

 [deps]
+ArraysOfArrays = "65a8f2f4-9b39-5baf-92e2-a9cc46fdf018"


I don't think the implementation has to or should use ArraysOfArrays explicitly - I guess you can just work with generic AbstractVector{<:AbstractVector}, and then one can e.g. provide such Array wrappers from ArraysOfArrays if desired.

Ok. But then there are some significant adjustments I have to make, cause I've relied on ArraysOfArrrays to do stuff like convert matrices to vector of vectors. What do you suggest in order to dispatch them?

IMO we should remove the matrix dispatch (even regardless of ArraysOfArrays) as data can be organized in rows or columns - but this is not clear if just a matrix is provided. I think users should just use eachrow/eachcol (in a future Julia version or with Compat: JuliaLang/julia#32310), ArraysOfArrays or ColVecs/RowVecs in KernelFunctions (I wanted to wait with extracting them a bit since I was hoping for the Julia PR).

Ok, I'll drop the matrix dispatch.

I just realized that computations of weighted mean, var and cov depend on ArraysOfArrays. Do you know a workaround?

They all work for vectors of vectors (of course, ArraysOfArrays, ColVecs or EachCol may implement more optimized versions of mean etc., but the default fallback works always):

julia> using Statistics julia> x = [rand(5) for _ in 1:5]; julia> mean(x) 5-element Vector{Float64}: 0.2609279080845831 0.5184741034528969 0.523898105188404 0.6408166253497452 0.33381236831836236 julia> var(x) 5-element Vector{Float64}: 0.019751971071567172 0.07655721742732527 0.06221826760951234 0.022761714823869204 0.10639711437483416 julia> cov(x) 5×5 Matrix{Float64}: 0.019752 -0.0182541 -0.00331084 0.0142255 -0.0231928 -0.0182541 0.0765572 -0.0500437 -0.0402498 0.0677088 -0.00331084 -0.0500437 0.0622183 0.0214676 -0.0425375 0.0142255 -0.0402498 0.0214676 0.0227617 -0.0405306 -0.0231928 0.0677088 -0.0425375 -0.0405306 0.106397

Not the weighted version. I ended up writing a for-loop. Don't know if it's the most efficient way.

mean(x, weights(p))

does not work.

Ah I didn't realize that you need the weighted ones. They are defined in StatsBase, there are some issues about exactly the vector of vectors case: JuliaStats/StatsBase.jl#534 and JuliaStats/StatsBase.jl#518

Here is what I came up with:

function cov(d::MvDiscreteNonParametric) x = hcat(support(d)...) p = probs(d) return cov(x, Weights(p, one(eltype(p))), 2,corrected = false) end

@devmotion , are you fine with this implementation?

src/multivariate/mvdiscretenonparametric.jl

codecov-commenter · 2021-11-16T01:19:29Z

Codecov Report

Merging #1424 (288401b) into master (7670bed) will decrease coverage by 0.51%.
The diff coverage is 89.47%.

❗ Current head 288401b differs from pull request most recent head eeba4f2. Consider uploading reports for the commit eeba4f2 to get more accurate results

@@            Coverage Diff             @@
##           master    #1424      +/-   ##
==========================================
- Coverage   84.42%   83.91%   -0.52%     
==========================================
  Files         124      123       -1     
  Lines        7456     7038     -418     
==========================================
- Hits         6295     5906     -389     
+ Misses       1161     1132      -29

Impacted Files	Coverage Δ
src/multivariates.jl	`70.66% <45.45%> (+32.20%)`	⬆️
src/multivariate/mvdiscretenonparametric.jl	`100.00% <100.00%> (ø)`
src/samplers/multinomial.jl	`51.72% <0.00%> (-32.90%)`	⬇️
src/matrixvariates.jl	`76.81% <0.00%> (-18.43%)`	⬇️
src/common.jl	`66.66% <0.00%> (-12.93%)`	⬇️
src/utils.jl	`90.47% <0.00%> (-9.53%)`	⬇️
src/multivariate/dirichlet.jl	`64.89% <0.00%> (-7.17%)`	⬇️
src/univariate/continuous/studentizedrange.jl	`72.72% <0.00%> (-2.28%)`	⬇️
src/univariate/continuous/epanechnikov.jl	`58.33% <0.00%> (-2.20%)`	⬇️
src/univariate/continuous/noncentralchisq.jl	`78.94% <0.00%> (-2.01%)`	⬇️
... and 90 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7670bed...eeba4f2. Read the comment docs.

devmotion · 2021-11-16T01:21:00Z

src/multivariate/mvdiscretenonparametric.jl

+"""
+Base.size(d::MvDiscreteNonParametric) = size(flatview(d.support)')
+
+function _rand!(rng::AbstractRNG, d::MvDiscreteNonParametric, x::AbstractVector{T}) where T <: Real


Since this is basically the same as the univariate implementation, it would be good to merge them.

src/multivariate/mvdiscretenonparametric.jl

Co-authored-by: David Widmann <devmotion@users.noreply.github.com>

… into MvDiscreteNonParametric

davibarreira · 2021-11-20T00:43:11Z

@devmotion, can I run the format from the terminal here in the same way we do in OptimalTransport.jl? My editor sometimes "auto-formats" stuff -.- , so I end up messing the previous formatting.

devmotion · 2021-11-20T00:58:49Z

No, we don't use a specific style here (yet). (BTW in general running the JuliaFormatter won't necessarily revert all of your changes.)

davibarreira · 2021-11-20T01:51:02Z

No, we don't use a specific style here (yet). (BTW in general running the JuliaFormatter won't necessarily revert all of your changes.)

Thanks. I ended up restoring the originals and doing the changes again, avoiding the auto-formatting.

… ArraysOfArrays.

… into MvDiscreteNonParametric

…t working.

davibarreira · 2021-12-19T13:08:36Z

@devmotion , I've implemented some of your suggestions (I still have to adjust DiscreteNonParametric to be used with GeneralNonParametric), but I'm having a hard time to get the sampler to work for MvDiscreteNonParametric. The rand!() function simply does not work. I mean, when I do:

x = collect(eachrow(rand(10,3)))
p = normalize!(rand(10),1)
μ = MvDiscreteNonParametric(x,p)
rand(μ)

it's working. But if I do

rand(μ,1)

I get instead:

3×2 Matrix{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}:
 [0.0969373, 0.623776, 0.447577]  [0.824966, 0.174585, 0.805307]
 [0.824966, 0.174585, 0.805307]   [0.0969373, 0.623776, 0.447577]
 [0.562016, 0.439165, 0.146882]   [0.562016, 0.439165, 0.146882]

devmotion · 2021-12-19T22:58:24Z

The eltype definition is incorrect. It should return the element type of the samples for arrayvariate distributions (ie., e.g., Float64 if the samples are AbstractVector{Float64}) but currently in the multivariate case you define it as a vector.

davibarreira · 2021-12-20T11:55:24Z

I've modified the code to

Base.eltype(::Type{<:MvDiscreteNonParametric{T}}) where T = Base.eltype(T)

Now, this returns for example Float64, but the rand(mu, 1) now does not work. I get an errror:
DimensionMismatch("inconsistent array dimensions")

devmotion

The definition of size is not correct, it can just be removed.

devmotion · 2021-12-20T15:59:33Z

src/multivariate/mvdiscretenonparametric.jl

+
+
+Base.length(d::MvDiscreteNonParametric) = length(first(d.support))
+Base.size(d::MvDiscreteNonParametric) = (length(d), length(d.support))


This is wrong and defined automatically based on length:

Suggested change

Base.size(d::MvDiscreteNonParametric) = (length(d), length(d.support))

Ok, done. But I'm still getting the erro :/, but instead it's:

StackOverflowError: Stacktrace: [1] rand!(rng::Random._GLOBAL_RNG, s::MvDiscreteNonParametric{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}, Float64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}, Vector{Float64}}, x::SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}) @ Distributions ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:83 [2] _rand!(rng::Random._GLOBAL_RNG, s::MvDiscreteNonParametric{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}, Float64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}, Vector{Float64}}, x::SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}) @ Distributions ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:117 [3] rand!(rng::Random._GLOBAL_RNG, s::MvDiscreteNonParametric{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}, Float64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}, Vector{Float64}}, x::SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}) @ Distributions ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:91 --- the last 2 lines are repeated 14426 more times --- [28856] _rand!(rng::Random._GLOBAL_RNG, s::MvDiscreteNonParametric{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}, Float64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}, Vector{Float64}}, x::SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}) @ Distributions ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:117 [28857] rand! @ ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:91 [inlined] [28858] _rand!(rng::Random._GLOBAL_RNG, s::MvDiscreteNonParametric{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}, Float64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}, Vector{Float64}}, x::Matrix{Float64}) @ Distributions ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:117 [28859] rand! @ ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:108 [inlined] [28860] rand @ ~/MEGA/EMAp/Distributions.jl/src/multivariates.jl:21 [inlined] [28861] rand(s::MvDiscreteNonParametric{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}, Float64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Int64, Base.Slice{Base.OneTo{Int64}}}, true}}, Vector{Float64}}, dims::Int64) @ Distributions ~/MEGA/EMAp/Distributions.jl/src/genericrand.jl:22 [28862] top-level scope @ In[31]:1 [28863] eval @ ./boot.jl:373 [inlined]

What did you run? And did you use the code in this PR with only size removed? Or did you make any additional changes?

I had added some changes. The _rand! was missing. I added it back, and it worked. Tks

davibarreira · 2021-12-21T12:07:47Z

@devmotion , this new commit has possibly all your corrections with the exception of incorporating the DiscreteNonParametric inside the GeneralNonParametric. The MvDiscreteNonParametric should be working. Should we do the modifications of DiscreteNonParametric in this PR?

devmotion · 2021-12-21T13:08:53Z

src/Distributions.jl

@@ -98,6 +98,7 @@ export
    FullNormalCanon,
    Gamma,
    DiscreteNonParametric,
+    GeneralDiscreteNonParametric,


Maybe let's keep it unexported until everything is figured out?

Suggested change

GeneralDiscreteNonParametric,

devmotion · 2021-12-21T13:13:52Z

src/multivariate/generaldiscretenonparametric.jl

+function rand(rng::AbstractRNG, d::GeneralDiscreteNonParametric)
+    x = support(d)
+    p = probs(d)
+    n = length(p)
+    draw = rand(rng, float(eltype(p)))
+    cp = p[1]
+    i = 1
+    while cp <= draw && i < n
+        @inbounds cp += p[i +=1]
+    end
+    return x[i]
+end


While this method is correct, it messes with the rand dispatches. Maybe either change it to the univariate case only (i.e., use ::GeneralDiscreteNonParametric{Univariate})

Suggested change

function rand(rng::AbstractRNG, d::GeneralDiscreteNonParametric)

x = support(d)

p = probs(d)

n = length(p)

draw = rand(rng, float(eltype(p)))

cp = p[1]

i = 1

while cp <= draw && i < n

@inbounds cp += p[i +=1]

end

return x[i]

end

function rand(rng::AbstractRNG, d::GeneralDiscreteNonParametric{Univariate})

x = support(d)

p = probs(d)

n = length(p)

draw = rand(rng, float(eltype(p)))

cp = p[1]

i = 1

while cp <= draw && i < n

@inbounds cp += p[i +=1]

end

return x[i]

end

or let's remove it for now and add it in a follow-up PR that updates DiscreteNonParametric:

Suggested change

function rand(rng::AbstractRNG, d::GeneralDiscreteNonParametric)

x = support(d)

p = probs(d)

n = length(p)

draw = rand(rng, float(eltype(p)))

cp = p[1]

i = 1

while cp <= draw && i < n

@inbounds cp += p[i +=1]

end

return x[i]

end

devmotion · 2021-12-21T13:14:49Z

src/multivariate/generaldiscretenonparametric.jl

+"""
+    support(d::MvDiscreteNonParametric)
+Get a sorted AbstractVector defining the support of `d`.
+"""


No need for a docstring, support is already documented (and I don't think it is sorted here actually?)

Suggested change

"""

support(d::MvDiscreteNonParametric)

Get a sorted AbstractVector defining the support of `d`.

"""

devmotion · 2021-12-21T13:15:38Z

src/multivariate/generaldiscretenonparametric.jl

+    probs(d::MvDiscreteNonParametric)
+Get the vector of probabilities associated with the support of `d`.


Suggested change

probs(d::MvDiscreteNonParametric)

Get the vector of probabilities associated with the support of `d`.

probs(d::GeneralDiscreteNonParametric)

Return the vector of probabilities associated with the support of `d`.

devmotion · 2021-12-21T13:18:22Z

src/multivariate/generaldiscretenonparametric.jl

+
+Base.length(d::GeneralDiscreteNonParametric) = length(first(d.support))


length should be defined for the multivariate case only, for other cases we need size:

Suggested change

Base.length(d::GeneralDiscreteNonParametric) = length(first(d.support))

Base.length(d::GeneralDiscreteNonParametric{Multivariate}) = length(first(d.support))

Base.size(d::GeneralDiscreteNonParametric{<:ArrayLikeVariate}) = size(first(d.support))

IMO it is better to be a bit conservative here and only define it for ArrayLikeVariates.

devmotion · 2021-12-21T13:32:44Z

src/multivariate/mvdiscretenonparametric.jl

+```
+"""
+function MvDiscreteNonParametric(
+    support::AbstractArray{<:AbstractVector{<:Real}},


This should be

Suggested change

support::AbstractArray{<:AbstractVector{<:Real}},

support::AbstractVector{<:AbstractVector{<:Real}},

in the multivariate case, and

Suggested change

support::AbstractArray{<:AbstractVector{<:Real}},

support::AbstractVector{<:AbstractArray{<:Real,N}},

only in the more general case of ArrayLikeVariate{N}.

Should I just use the second option and change the function to GeneralDiscreteNonParametric?

devmotion · 2021-12-21T13:33:01Z

src/multivariate/mvdiscretenonparametric.jl

+"""
+function MvDiscreteNonParametric(
+    support::AbstractArray{<:AbstractVector{<:Real}},
+    p::AbstractVector{<:Real} = fill(inv(length(support)), length(support)),


I'm not sure if we should add a default definition for p. In any case, we should use the non-allocating Fill instead.

Suggested change

p::AbstractVector{<:Real} = fill(inv(length(support)), length(support)),

p::AbstractVector{<:Real} = Fill(inv(length(support)), length(support)),

devmotion · 2021-12-21T13:34:07Z

src/multivariate/mvdiscretenonparametric.jl

+    )
+end
+
+Base.eltype(::Type{<:MvDiscreteNonParametric{T}}) where T = Base.eltype(T)


This should be defined more generally:

Suggested change

Base.eltype(::Type{<:MvDiscreteNonParametric{T}}) where T = Base.eltype(T)

Base.eltype(::Type{<:GeneralDiscreteNonParametric{<:ArrayLikeVariate,T}}) where T = eltype(T)

devmotion · 2021-12-21T13:39:35Z

src/multivariate/mvdiscretenonparametric.jl

+function mean(d::MvDiscreteNonParametric)
+    return StatsBase.mean(hcat(d.support...), Weights(d.p, one(eltype(d.p))),dims=2)
+end
+
+function var(d::MvDiscreteNonParametric)
+    x = hcat(support(d)...)
+    p = probs(d)
+    return StatsBase.var(x, Weights(p, one(eltype(p))), 2,corrected = false)
+end
+
+function cov(d::MvDiscreteNonParametric)
+    x = hcat(support(d)...)
+    p = probs(d)
+    return cov(x, Weights(p, one(eltype(p))), 2,corrected = false)
+end


I requested to add support for vectors of vectors to StatsBase but it might take a bit until it is available. IMO it is fine to have possibly less optimized implementations until then. However, we should avoid splatting and use

Suggested change

function mean(d::MvDiscreteNonParametric)

return StatsBase.mean(hcat(d.support...), Weights(d.p, one(eltype(d.p))),dims=2)

end

function var(d::MvDiscreteNonParametric)

x = hcat(support(d)...)

p = probs(d)

return StatsBase.var(x, Weights(p, one(eltype(p))), 2,corrected = false)

end

function cov(d::MvDiscreteNonParametric)

x = hcat(support(d)...)

p = probs(d)

return cov(x, Weights(p, one(eltype(p))), 2,corrected = false)

end

function mean(d::MvDiscreteNonParametric)

return mean(reduce(hcat, d.support), Weights(d.p, one(eltype(d.p))); dims=2)

end

function var(d::MvDiscreteNonParametric)

x = reduce(hcat, support(d))

p = probs(d)

return var(x, Weights(p, one(eltype(p))), 2; corrected = false)

end

function cov(d::MvDiscreteNonParametric)

x = reduce(hcat, support(d))

p = probs(d)

return cov(x, Weights(p, one(eltype(p))), 2; corrected = false)

end

devmotion · 2021-12-21T13:40:13Z

src/univariate/discrete/discretenonparametric.jl

-function rand(rng::AbstractRNG, d::DiscreteNonParametric)
-    x = support(d)
-    p = probs(d)
-    n = length(p)
-    draw = rand(rng, float(eltype(p)))
-    cp = p[1]
-    i = 1
-    while cp <= draw && i < n
-        @inbounds cp += p[i +=1]
-    end
-    return x[i]
-end
-


I guess this was removed accidentally? I guess it would be cleaner to not make any changes to DiscreteNonParametric in this PR.

yeah. my bad

davibarreira · 2023-07-24T21:21:04Z

@juliohm , this might interest you.

davibarreira added 10 commits November 15, 2021 20:55

Implementation of MvDiscreteNonParametric distribution.

34181c4

Implementation of MvDiscreteNonParametric samplers.

0cc7518

Added ArraysOfArrays as dependency.

67b1890

Added MvDiscreteNonParametric to be exported.

7fba9da

Added MvDiscreteNonParametric to be exported.

6f43b18

Added MvDiscreteNonParametric test set.

c64339a

Added MvDiscreteNonParametric samplers to samplers.jl.

e156052

Added _rand!(::GlobalRNG,...).

d11d300

Moved MvDiscreteNonParametric sampling to just one file. All tests pa…

24612da

…ssing.

Merge branch 'master' into MvDiscreteNonParametric

288401b

devmotion reviewed Nov 16, 2021

View reviewed changes

src/multivariate/mvdiscretenonparametric.jl Outdated Show resolved Hide resolved

devmotion reviewed Nov 16, 2021

View reviewed changes

src/multivariate/mvdiscretenonparametric.jl Outdated Show resolved Hide resolved

devmotion reviewed Nov 16, 2021

View reviewed changes

src/multivariate/mvdiscretenonparametric.jl Outdated Show resolved Hide resolved

davibarreira and others added 5 commits November 16, 2021 20:50

Apply suggestions from code review

e8cf084

Co-authored-by: David Widmann <devmotion@users.noreply.github.com>

Merge branch 'master' of https://github.com/JuliaStats/Distributions.jl…

68c4d4d

… into MvDiscreteNonParametric

Removing ArraysOfArarys dependency.

cbb6ae7

ArrarysOfArrays on test.

67a15dd

Fixed GeneralDiscreteNonParametric with version from devmotion.

5644219

Reverting wrong formatting changes.

3baece6

davibarreira added 3 commits November 19, 2021 23:40

Fixed functionalities mean, var and cov for vector of vectors without…

10ba493

… ArraysOfArrays.

Removing testing.

236380a

Merge branch 'master' of https://github.com/JuliaStats/Distributions.jl…

2947ca9

… into MvDiscreteNonParametric

Added rand function to generaldiscretenonparametric, but _rand! is no…

eeba4f2

…t working.

devmotion mentioned this pull request Dec 19, 2021

Wrong weighted sum and mean results for complex numbers JuliaStats/StatsBase.jl#518

Closed

devmotion reviewed Dec 20, 2021

View reviewed changes

davibarreira added 4 commits December 20, 2021 20:11

Moving some functions to generaldiscretenonparametric.

b208df3

Moving some functions to generaldiscretenonparametric.

b2f0ec0

Removing ArraysOfArarys as dependency on tests.

ad6d01e

Renamed test set.

001d73d

devmotion reviewed Dec 21, 2021

View reviewed changes

		@@ -0,0 +1,153 @@
		struct MvDiscreteNonParametric{T <: Real,P <: Real,Ts <: ArrayOfSimilarArrays{T},Ps <: AbstractVector{P}} <: DiscreteMultivariateDistribution

	struct MvDiscreteNonParametric{T <: Real,P <: Real,Ts <: ArrayOfSimilarArrays{T},Ps <: AbstractVector{P}} <: DiscreteMultivariateDistribution
	struct GeneralDiscreteNonParametric{VF,T,P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} <: Distribution{VF,Discrete}

	struct MvDiscreteNonParametric{T <: Real,P <: Real,Ts <: ArrayOfSimilarArrays{T},Ps <: AbstractVector{P}} <: DiscreteMultivariateDistribution
	const MvDiscreteNonParametric{T<:AbstractVector{<:Real},P<:Real,Ts<:AbstractVector{T},Ps<:AbstractVector{P}} = GeneralDiscreteNonParametric{Multivariate,T,P,Ts,Ps}

		for i in Base.OneTo(size(m, 2))
		_rand!(rng, smp, view(m, :, i))



		Base.length(d::MvDiscreteNonParametric) = length(first(d.support))
		Base.size(d::MvDiscreteNonParametric) = (length(d), length(d.support))

		probs(d::MvDiscreteNonParametric)
		Get the vector of probabilities associated with the support of `d`.


		Base.length(d::GeneralDiscreteNonParametric) = length(first(d.support))

-Base.length(d::GeneralDiscreteNonParametric) = length(first(d.support))
+Base.length(d::GeneralDiscreteNonParametric{Multivariate}) = length(first(d.support))
+Base.size(d::GeneralDiscreteNonParametric{<:ArrayLikeVariate}) = size(first(d.support))

	support::AbstractArray{<:AbstractVector{<:Real}},
	support::AbstractVector{<:AbstractVector{<:Real}},

	support::AbstractArray{<:AbstractVector{<:Real}},
	support::AbstractVector{<:AbstractArray{<:Real,N}},

	p::AbstractVector{<:Real} = fill(inv(length(support)), length(support)),
	p::AbstractVector{<:Real} = Fill(inv(length(support)), length(support)),

	Base.eltype(::Type{<:MvDiscreteNonParametric{T}}) where T = Base.eltype(T)
	Base.eltype(::Type{<:GeneralDiscreteNonParametric{<:ArrayLikeVariate,T}}) where T = eltype(T)

New Distribution: MvDiscreteNonParametric #1424

Are you sure you want to change the base?

New Distribution: MvDiscreteNonParametric #1424

Conversation

davibarreira commented Nov 16, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davibarreira Nov 23, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov-commenter commented Nov 16, 2021 • edited Loading

Codecov Report

Choose a reason for hiding this comment

davibarreira commented Nov 20, 2021

devmotion commented Nov 20, 2021

davibarreira commented Nov 20, 2021

davibarreira commented Dec 19, 2021

devmotion commented Dec 19, 2021

davibarreira commented Dec 20, 2021

devmotion left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davibarreira Dec 20, 2021 • edited Loading

Choose a reason for hiding this comment

davibarreira commented Dec 21, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davibarreira commented Jul 24, 2023

davibarreira Nov 23, 2021 •

edited

Loading

codecov-commenter commented Nov 16, 2021 •

edited

Loading

davibarreira Dec 20, 2021 •

edited

Loading

davibarreira commented Dec 21, 2021 •

edited

Loading