[Bug] RafteryCopula fixes.

[lrnv]

Fixes #98

@Santymax98 it loks like the pdf is now fixed, but the cdf has values that still do not match what is in the tests files (see the @test_broken).

two month ago there #98 (comment) you proposed to use

function prueba(R::Vector{T}, u::Vector{T}) where T
    # Order the vector u
    u_ordered = sort(u)
    println("Sorted vector: ", u_ordered)
    
    term1 = u_ordered[1]
    println("Term 1: ", term1)
    
    term2 = ((1 - R[1]) * (1 - R[2])) / (1 - R[1] - R[2]) * prod(u)^(1/(1 - R[1]))
    println("Term 2: ", term2)
    
    term3 = 0.0
    for i in 2:length(R) # <<<<<<<<--------------- This should be 2:d and not 2:length(R) since length(R) is not the dimension. 
        prod_prev = prod(u_ordered[1:i-1])
        term3_part = (R[1] * (1 - R[1])) / ((1 - R[1] - i) * (2 - R[1] - i)) * prod_prev^(1/(1 - R[1])) * u_ordered[i]^((2 - R[1] - i) / (1 - R[1]))
        println("Term 3 (part $i): ", term3_part)
        term3 += term3_part
    end
    
    # Combine the terms to get the cumulative distribution function
    cdf_value = term1 + term2 - term3
    println("Final CDF value: ", cdf_value)
    
    return cdf_value
end

prueba([0.5,3], [0.1,0.2,0.3])

to check that the value's returned by the code are indeed correct. But your code has a mistake.

It looks to me like the current implementation is mathcing what is in the article, so maybe the issue is that the tests values are now wrong and should be updated ?

[Santymax98]

I'm going to read it, in the morning I'm working but later I'll be free and I hope to review the tests. I remember that I did the tests with paper and pencil according to the formula in the document. El El lun, 12 feb 2024 a la(s) 5:20, Oskar Laverny ***@***.***> escribió:

…

Fixes #98 <#98> @Santymax98 <https://github.com/Santymax98> it loks like the pdf is now fixed, but the cdf has values that still do not match what is in the tests files (see the @test_broken). two month ago there #98 (comment) <#98 (comment)> you proposed to use function prueba(R::Vector{T}, u::Vector{T}) where T # Order the vector u u_ordered = sort(u) println("Sorted vector: ", u_ordered) term1 = u_ordered[1] println("Term 1: ", term1) term2 = ((1 - R[1]) * (1 - R[2])) / (1 - R[1] - R[2]) * prod(u)^(1/(1 - R[1])) println("Term 2: ", term2) term3 = 0.0 for i in 2:length(R) # <<<<<<<<--------------- This should be 2:d and not 2:length(R) since length(R) is not the dimension. prod_prev = prod(u_ordered[1:i-1]) term3_part = (R[1] * (1 - R[1])) / ((1 - R[1] - i) * (2 - R[1] - i)) * prod_prev^(1/(1 - R[1])) * u_ordered[i]^((2 - R[1] - i) / (1 - R[1])) println("Term 3 (part $i): ", term3_part) term3 += term3_part end # Combine the terms to get the cumulative distribution function cdf_value = term1 + term2 - term3 println("Final CDF value: ", cdf_value) return cdf_valueend prueba([0.5,3], [0.1,0.2,0.3]) to check that the value's returned by the code are indeed correct. But your code has a mistake. It looks to me like the current implementation is mathcing what is in the article, so maybe the issue is that the tests values are now wrong and should be updated ? ------------------------------ You can view, comment on, or merge this pull request online at: #137 Commit Summary - 75b06de <75b06de> Fixing the pdf and clearing out the code. File Changes (2 files <https://github.com/lrnv/Copulas.jl/pull/137/files>) - *M* src/MiscellaneousCopulas/RafteryCopula.jl <https://github.com/lrnv/Copulas.jl/pull/137/files#diff-683b121951621fad248976c0f09fa07e079e95245b4a504be601475a7293e307> (49) - *M* test/RafteryTest.jl <https://github.com/lrnv/Copulas.jl/pull/137/files#diff-1fac2c1637f27ccc8e7b152e540c588b4b3e4f85668273eac3b05cbb8bbc70bb> (37) Patch Links: - https://github.com/lrnv/Copulas.jl/pull/137.patch - https://github.com/lrnv/Copulas.jl/pull/137.diff — Reply to this email directly, view it on GitHub <#137>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A7PTWNUNEEULBGBL42NEGJDYTHUGZAVCNFSM6AAAAABDEOZV7KVHI2DSMVQWIX3LMV43ASLTON2WKOZSGEZDSOBRHA4DCOI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[lrnv]

I remember that I did the tests with paper and pencil according to the formula in the document.

This is indeed a very good way of insuring that the implementation is OK. Maybe you made a mistake doing this, could you simply re-do it for the trivariate cdf tests ? the only ones that are still broken.

[Santymax98]

function prueba_CDF(R::Vector{T}, u::Vector{T}) where T
    # Order the vector u
    θ = R[1]
    println("param:", θ)
    d = round(Int, R[2])
    println("dimension:", d)
    u_ordered = sort(u)
    println("Sorted vector: ", u_ordered)
    
    term1 = u_ordered[1]
    println("Term 1: ", term1)
    
    term2 = ((1 - θ) * (1 -d)) / (1 - θ - d) * prod(u)^(1/(1 - θ))
    println("Term 2: ", term2)
    
    term3 = 0.0
    for i in 2:d # <<<<<<<<--------------- This should be 2:d and not 2:length(R) since length(R) is not the dimension. 
        prod_prev = prod(u_ordered[1:i-1])
        term3_part = ((θ * (1 - θ)) / ((1 - θ - i) * (2 - θ - i))) * prod_prev^(1/(1 - θ)) * u_ordered[i]^((2 - θ - i) / (1 - θ))
        println("Term 3 (part $i): ", term3_part)
        term3 += term3_part
    end
    
    # Combine the terms to get the cumulative distribution function
    cdf_value = term1 + term2 - term3
    println("Final CDF value: ", cdf_value)
    
    return cdf_value
end

prueba_CDF([0.5,3], [0.1,0.2,0.3])

should be 0.08236... I calculated this value manually and the test code works fine. However, when I combine it with Copulas.jl it gives me the error that we have had so far.

[Santymax98]

function prueba_PDF(R::Vector{T}, u::Vector{T}) where T
    # Order the vector u
    θ = R[1]
    d = round(Int, R[2])
    u_ordered = sort(u)
    println("Sorted vector: ", u_ordered)
    
    term1 = (1/(((1-θ)^(d-1))*(1-θ-d)))
    println("Term 1: ", term1)
    
    term2 = (1-d-θ*(u_ordered[d])^((1-θ-d)/(1-θ)))
    println("Term 2: ", term2)
    
    term3 = (prod(u))^((θ)/(1-θ))
    println(term3)
    # Combine the terms to get the density distribution function
    pdf_value = term1*term2*term3
    println("Final PDF value: ", pdf_value)
    
    return pdf_value
end
prueba_PDF([0.5,3], [0.1,0.2,0.3])

Regarding this test, the value in this case should be 1.99450... and on my machine the results are coming out correct.
Do you think that regarding density there is any problem with calculating the logarithm? Is it simpler to do it with multiplication?

[lrnv]

@Santymax98 I cleared up the code and fixed the previous value. I included your versions as fallbak testing. Thanks a lot this is now Ok for me !