How to add dimensionless constraints - Π theorem

[Alexia369]

Hello, how can I use the custom loss function to
(1) obtain the dimension exponential vector of the variable
(2) If I can obtain it, I can add the dimensionless constraint -Π theorem to get the desired dimensionless parameter
I am not very good at writing Julia code, and I always get an error when using Unitful.Dimensions()
I would like to ask you how to do it, thank you

[Alexia369]

dimensionless_constants_only= True
I wonder if this line of code has converged to or found a trend for dimensionless parameters，i guess

[MilesCranmer]

Yep that’s the parameter. Does that answer the problem or did you still want to use a custom loss?

[MilesCranmer]

This is where the dimensionality check occurs in the default loss function: https://github.com/MilesCranmer/SymbolicRegression.jl/blob/71447ee7bee18fc836a7d7bb4c8ce5d249bc0e7c/src/LossFunctions.jl#L71

And this is where dimensional constraint violations are calculated: https://github.com/MilesCranmer/SymbolicRegression.jl/blob/71447ee7bee18fc836a7d7bb4c8ce5d249bc0e7c/src/DimensionalAnalysis.jl#L195

Note that it uses DynamicQuantities; not Unitful

[Alexia369]

I referred to the DynamicQuantities.jl document you provided, which mentioned that umass ulength can obtain the dimensional index, but it still reports an error when actually used. Why is that?

julia> umass(x)
1//1
julia> ulength(x)
1//2

dim = [DynamicQuantities.ulength(s[1]), DynamicQuantities.utime(s[1])]
Error reason: MethodError: no method matching ulength(::Float32)
Can you please guide me on how to write this Julia code? This is very important to me. Thank you very much

[Alexia369]

Yes，I still want to use a custom loss

[MilesCranmer]

X does not have units; they need to be multiplied in from X_units. And DynamicQuantities is not imported automatically in the loss function, you would need to do that separately.