Crash in visualization

[claudiocc1]

hi,

I went back to an old project. It used to work with the then-current version of iminuit. Sorry, cannnot remember the exact version. Now it crashes with iminuit 2.25 (python 3.11.5, Intel Mac.).

The fit actually works, and gives the exact same result now as it used to. I think the crash is in the visualization part, which I guess has been added since?

Here you can find my jupyter notebook, stripped down to the bare minimum (html and ipynb versions) http://hep.ucsb.edu/people/claudio/minuit/

Thanks.

Claudio

PS is there a way to turn off the visualization?

[HDembinski]

Thanks for reporting, that is indeed a bug. I will see how this can be fixed. A temporary workaround for you is

# Minuit fit
p0 = np.array([2., 2., 0., -1., -1.5])
mFit = LeastSquares(x, y, ey, fitFun)
mFit.visualize = lambda x, *args: None
m = Minuit(mFit, p0)  # pass starting values
m.migrad()
# m.hesse()

[HDembinski]

If you want to fit a smooth function $f(x)$ through your data points under the condition $f(x) \ge 0$, you can also use Bernstein polynomials. They are guaranteed to be positive inside the domain, if all the parameters are positive. You cannot safely extrapolate Bernstein polynomials outside their domain, though. If you need to query the function outside the fitted range, make the domain of the polynomial as large as you need to query it.

from numba_stats import bernstein

def model(x, p):
    domain_of_polynomial = (-0.4, 0.4)
    return bernstein.density(x, p, *domain_of_polynomial)

start = np.ones(4)  # increase this to increase flexibility of Bernstein polynomial
m = Minuit(LeastSquares(x, y, ey, model), start) 
m.limits = (0, None)  # ensure positiveness
m.migrad()

[claudiocc1]

Yes I know :) We are working with people that like splines (I dont), so here we are.... C.

[HDembinski]

I like splines, too :). Bernstein polynomials are a special case of splines.