DIVAnd.fit issue about divand.jl

[ctroupin]

hello, can you provide us with a minimum working example (i.e. subset of the dataset) and the commands you ran to get this issue?
Otherwise the error messages don't help too much in this case, I'm afraid.

[fricour]

Data : divand_github_issue.txt

Hello,

The subset is given in the attached txt file above (columns order : lon, lat, sigma (density) and depth).
If we create a matrix m with lon, lat, sigma and depth (size of 64800x4):

This command works : ~, len, ~,~,~ = DIVAnd.fit_isotropic((m[:,1],m[:,2]), m[:,3], collect(0:1:20), 100)
but this one does not ~, len, ~,~,~ = DIVAnd.fit((m[:,1],m[:,2]), m[:,3], collect(0:1:20), 100)

I also added the depth field in the text file because I was testing the function fithorzlen, which gives me an error as well. I used the following command :

lenz, dbinfo = DIVAnd.fithorzlen((m[:,1],m[:,2],m[:,4]), m[:,3], [50.,60,70]) where I used a variable z equal to [50.,60,70] (´z´ is not defined in the documentation, I had to go in one of the examples to see what it was related to anf got the following result:

┌ Warning: Be patient big data set: 64800

└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:675
┌ Warning: Too few data. Will use guesses (np = 0, RLz = 1.811113649598692, )
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:894
┌ Info: Data points at z=50.0: 64800, horz. correlation length: 1.811113649598692
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:1061
┌ Warning: Be patient big data set: 64800
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:675
┌ Warning: Too few data. Will use guesses (np = 0, RLz = 1.811113649598692, )
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:894
┌ Info: Data points at z=60.0: 64800, horz. correlation length: 1.811113649598692
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:1061
┌ Warning: Be patient big data set: 64800
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:675
┌ Warning: Too few data. Will use guesses (np = 0, RLz = 1.811113649598692, )
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:894
┌ Info: Data points at z=70.0: 64800, horz. correlation length: 1.811113649598692
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:1061

All elements in A are equal to the fill-value. No filling is possible

Stacktrace:
[1] error(::String) at ./error.jl:33
[2] DIVAnd_fill!(::Array{Float64,1}, ::Float64) at /home/flo/.julia/packages/DIVAnd/azb67/src/DIVAnd_fill!.jl:84
[3] #fithorzlen#522(::Float64, ::Float64, ::Float64, ::Float64, ::Function, ::Function, ::getfield(DIVAnd, Symbol("##526#533")), ::Int64, ::Bool, ::Array{Float64,1}, ::typeof(fithorzlen), ::Tuple{Array{Float64,1},Array{Float64,1},Array{Float64,1}}, ::Array{Float64,1}, ::Array{Float64,1}) at /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:1065
[4] fithorzlen(::Tuple{Array{Float64,1},Array{Float64,1},Array{Float64,1}}, ::Array{Float64,1}, ::Array{Float64,1}) at /home/flo/.julia/packages/DIVAnd/azb67/src/fit.jl:1022
[5] top-level scope at In[314]:1

Maybe I am using it all wrong but I am not sure ! I also fixed some bugs by updating julia from a version 1.0.x to 1.1.1 (e.g. DIVAndrun was not working on my laptop with julia version 1.0.x)

For information, all tests are OK when using Pkg.test("DIVAnd")

Thanks for your help!

[ctroupin]

Thanks for the files, that really help to try and reproduce the issue.

It seems the package NLopt is missing from the requirement list. While we are fixing this, you can add install NLopt if it is not yet there, then add the following line in DIVAnd.jl

using NLopt

then re-run the code with DIVAnd.fit(...). With your data I get len = [0.5, 0.5]

@Alexander-Barth : is it suffient to add NLopt in REQUIRE and using NLopt in DIVAnd.jl ?

[fricour]

Hello,

I installed NLopt and I did Pkg.update() in any case. I've rerun everything, have restarted the kernel, I still have the same issue.

[ctroupin]

And did you add using NLopt in DIVAnd.jl?

[fricour]

Yes of course !

[Alexander-Barth]

The functionality of DIVAnd.fit and DIVAnd.fit_isotropic have been replaced by other functions. In future version DIVAnd.fit and DIVAnd.fit_isotropic will be removed. Can you use the other function (DIVAnd.fithorzlen and DIVAnd.fitvertlen ) instead?

[fricour]

I did. I also got an issue (see also previous messages).

The error I get with DIVAnd.fithorzlen is, when using lenz, dbinfo = DIVAnd.fithorzlen((m[:,1],m[:,2],m[:,4]), m[:,3], [50.,60,70])

┌ Warning: Be patient big data set: 64800
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:675
┌ Warning: Too few data. Will use guesses (np = 0, RLz = 1.811113649598692, )
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:894
┌ Info: Data points at z=50.0: 64800, horz. correlation length: 1.811113649598692
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:1061
┌ Warning: Be patient big data set: 64800
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:675
┌ Warning: Too few data. Will use guesses (np = 0, RLz = 1.811113649598692, )
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:894
┌ Info: Data points at z=60.0: 64800, horz. correlation length: 1.811113649598692
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:1061
┌ Warning: Be patient big data set: 64800
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:675
┌ Warning: Too few data. Will use guesses (np = 0, RLz = 1.811113649598692, )
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:894
┌ Info: Data points at z=70.0: 64800, horz. correlation length: 1.811113649598692
└ @ DIVAnd /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:1061

All elements in A are equal to the fill-value. No filling is possible

Stacktrace:
[1] error(::String) at ./error.jl:33
[2] DIVAnd_fill!(::Array{Float64,1}, ::Float64) at /home/flo/.julia/packages/DIVAnd/lgTG2/src/DIVAnd_fill!.jl:84
[3] #fithorzlen#522(::Float64, ::Float64, ::Float64, ::Float64, ::Function, ::Function, ::getfield(DIVAnd, Symbol("##526#533")), ::Int64, ::Bool, ::Array{Float64,1}, ::typeof(fithorzlen), ::Tuple{Array{Float64,1},Array{Float64,1},Array{Float64,1}}, ::Array{Float64,1}, ::Array{Float64,1}) at /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:1065
[4] fithorzlen(::Tuple{Array{Float64,1},Array{Float64,1},Array{Float64,1}}, ::Array{Float64,1}, ::Array{Float64,1}) at /home/flo/.julia/packages/DIVAnd/lgTG2/src/fit.jl:1022
[5] top-level scope at In[31]:1

There seems to be something with the function fit at line 1022 so if the same function fit gives the error with DIVAnd.fit, it is still present in DIVAnd.fithorzlen maybe?

[Alexander-Barth]

When I remove the NaN values, it seems to work for me:

using DIVAnd, DelimitedFiles
m = readdlm("divand_github_issue.txt")
sel = findall(isfinite.(m[:,3]));
lenz, dbinfo = DIVAnd.fithorzlen((m[sel,1],m[sel,2],m[sel,4]), m[sel,3], [50.,60.,70.])
extrema(m[sel,4])
# returns 
# (61.11283874511719, 61.11283874511719)

Note that the depth level is are constant in this dataset.

And thanks, I added a description for the z parameter DIVAnd.fithorzlen.

[fricour]

Perfect, it now works ! The NaN values were the source of the issue then.

Thank you !

[Alexander-Barth]

Thank you for your confirmation!