n_buffer has wrong dimensions

[pascalschulthess]

When I ran a fit of a ODE model with 2 equations and 3 parameters against data with 19 data points, I encountered the following error:

ERROR: LoadError: MethodError: no method matching mul!(::Vector{Float64}, ::LinearAlgebra.Transpose{Float64, Matrix{Float64}}, ::Matrix{Float64}, ::Bool, ::Bool)

This hinted at a dimension mismatch in the mul! method. After some debugging, I found that in levenberg_marquardt.jl , there’s line 80 creating a buffer:

n_buffer = Vector{T}(undef, n)

Checking size(n_buffer) = (3,)

Line 123, where my code fails with the aforementioned error, is:

mul!(n_buffer, transpose(J), value(df))

with size(transpose(J))=(3,19) and size(value(df)) = (19,1). Multiplying these should result in a size (3,1) and not (3,).
Thus, it’s also not possible to save the product to n_buffer.

I believe that this would be fixed by changing:

n_buffer = Matrix{T}(undef, n, 1)

[JakobAsslaender]

Without a minimal working example it is hard to tell, but my guess is that your model function returns a 3x1 matrix instead of a vector. Can you check if that is the case? An easy fix would in that case be to change

function mymodel(t,p)
     ...
     return x
end

to

function mymodel(t,p)
     ...
     return vec(x)
end

[pkofod]

Without a minimal working example it is hard to tell, but my guess is that your model function returns a 3x1 matrix instead of a vector. Can you check if that is the case? An easy fix would in that case be to change

I agree, it appears that you return a matrix, not a vector.

julia> size(rand(19))
(19,)

julia> size(rand(19,1))
(19, 1)