density with weights

[Nosferican]

Consider

using StatsBase, StatPlots
density(1:3) # Fine
density(1:3, weights = weights([0,500,1])) # No change
density(1:3, weights = [0,500,1]) # No change
density(vcat(1, ones(500), 2)) # Correct
density(1:3, weights([0,500,1])) # Not sure what this is

[mkborregaard]

Hi, when opening issues, could you please post runnable code (including all usings) and post the plot if it's the plot that is the concern here?
Here's an MWE and a plot based on this:

using StatsBase, StatPlots
plot(
       density(1:3), # Fine
       density(1:3, weights = weights([0,500,1])), # No change
       density(1:3, weights = [0,500,1]), # No change
       density(vcat(1, ones(500), 2)), # Correct
       density(1:3, weights([0,500,1])), # Not sure what this is
legend = false) 

So, your concern is that the weights argument to density appears to be ignored by StatPlots?

[Nosferican]

Will do. That is correct.

[cossio]

So, your concern is that the weights argument to density appears to be ignored by StatPlots?

Yes, I'm having the same issue and would definitely like to have a usable weights keyword on density with the same function as the weights option in histogram.

[mkborregaard]

StatsPlots uses KernelDensity.kde for the density plots. Though I can see a way of hacking support for weights directly here, a more sustainable option might be to request this on KernelDensity? Happy to forward the support here if they add it there.

[Nosferican]

Just to verify before opening an issue with KernelDensity, the following feature doesn't give you what you would need to adapt the code here? JuliaStats/KernelDensity.jl#26 It was the only issue/PR I found regarding weights in KernelDensity.

[mkborregaard]

Oh I see it does work, just wasn't documented (where I looked)! Let me have a look.

[Nosferican]

Perfect! It would still be good to open an issue there to have it documented if that is still the case.

[mkborregaard]

Does this give what you'd expect @Nosferican ? #232

[Nosferican]

I think for the first four plots it is what I would expect. 💯I am not sure about the last one. From the documentation, I don't know what would density(::AbstractVector, ::AbstractVector) or density(::AbstractVector, ::Weights) should do. Whatever it is currently doing, is beyond me. 🤷‍♂Maybe that one should give a MethodError. Alternatively,

density(x::AbstractVector, wts::Weights) = density(x, weights = wts)

would make sense and would be what I would argue for as an improvement over current behavior.

[mkborregaard]

I'd argued for keeping it as is on the PR

[Nosferican]

Thanks for the PR!