Allow non-vectorized model functions? about lsqfit.jl HOT 10 CLOSED

Maybe with julia v0.6 it is enough to exemplify the use of the new @. macro in the README example?

from lsqfit.jl.

Here's a reduced example that might reveal what's going on:

julia> x = ones(2)
2-element Array{Float64,1}:
 1.0
 1.0

julia> x^2
ERROR: MethodError: `*` has no method matching *(::Array{Float64,1}, ::Array{Float64,1})
...

from lsqfit.jl.

Yes, exactly that is it, but it is going internally on the curve_fit routine and I seem to not be able to make it go away. I actually have a simpler example (adjusting a degree 2 polynomial) and it seems to trigger the same problem. Maybe I am not understanding the syntaxis right...

 model(x,p)=p[3]+p[2]*x+p[1]*x*x
    ydata=map(x->log(x),yys)
    rara=curve_fit(model,xxs,ydata, [0,0,0])

returns the same error.

from lsqfit.jl.

@mlubin was just being nice. He was trying to tell you the bug is on your end. When you define the function, you need to use elemental operations like .* because curve_fit internally is evaluating the function for vector values.

Specifically, you need to write

model(x,p)=(1/(sqrt(2*pi)*p[1])) * exp(-( (x-p[2])/(sqrt(2)*p[1])) .^2)

Note the extra period before the power.

from lsqfit.jl.

Oh, I know he was nice, anyone who responds is doing good. Was I rude? in that case my apologies, probably is because I write a bit rough. Going back to the issue, yes, it seems to be that I don't understand the point syntaxis in Julia well. How is that p[1] is NOT an element of an array? I put the point in every operator now and it works, but I still don't understand what it is supposed to mean (well I have another error now...)

from lsqfit.jl.

No, you were not rude. Sorry if I implied that.

p[1] is an element of an array, but in the context of the evaluation of the model function, it is a scalar. The argument x of the function appears to be a scalar to you, but it is a vector inside curve_fit.

Here is a simple example.

f(x,p) = x*p[1] + p[2]
f(1,[2,3]) = 1*2 + 3 = 5
f([1,2,3],[2,3]) = [5,7,9]

You are thinking of the function like the second line, while inside curve_fit it is being called like in the third line.

from lsqfit.jl.

I had this problem, but I didn't understand how to use the point yet. I put in diferente places but the ERROR continuing coming. Someone can help or say where I can find a tutorial about that?
My code:

`model(t, p) = ((p[1]*p[2])/(sqrt((1/sqrt(p[3]p[4]))^2-(p[2]/(2p[3]))^2)*p[3]))*sin(sqrt((1/sqrt(p[3]p[4]))^2-(p[2]/(2p[3]))^2)*t)exp(-(p[2]/(2p[3]))*t)

fit = curve_fit(model, time, Volts, [Vm, R, L, C])`

from lsqfit.jl.

@CaBeOli, when your model is evaluated, the t is a Vector{Float64}. So, anywhere you have a t on both sides of a binary operator, you should use the element-wise syntax. In your case, this should be sufficient:

model(t, p) = ((p[1]*p[2])/(sqrt((1/sqrt(p[3]p[4]))^2 -
              (p[2]/(2p[3]))^2)*p[3])) * sin(sqrt((1/sqrt(p[3]p[4]))^2-(p[2]/(2p[3]))^2)*t)
                                      .* exp(-(p[2]/(2p[3]))*t)

from lsqfit.jl.

Maybe the more appropriate question if whether we should modify LsqFit.jl to not require the user to write their model in a vectorized form?

from lsqfit.jl.

Maybe with julia v0.6 it is enough to exemplify the use of the new @. macro in the README example?

I agree.

from lsqfit.jl.