least_squares fit using migrad is broken

[s-kuberski]

import pyerrors as pe
import numpy as np
x = np.arange(10)
y = [pe.pseudo_Obs(i+1, i*.2+.01, 'test') for i in x]
[o.gamma_method() for o in y]
def func(a, x):
    return a[0] + a[1] * x
pe.fits.least_squares(x, y, func)
pe.fits.least_squares(x, y, func, method='migrad')

results in

KeyError                                  Traceback (most recent call last)
/opt/anaconda3/lib/python3.6/site-packages/scipy/optimize/optimize.py in __getattr__(self, name)
    121         try:
--> 122             return self[name]
    123         except KeyError:

KeyError: 'nit'

During handling of the above exception, another exception occurred:

AttributeError                            Traceback (most recent call last)
<ipython-input-18-79beea198971> in <module>
      7     return a[0] + a[1] * x
      8 pe.fits.least_squares(x, y, func)
----> 9 pe.fits.least_squares(x, y, func, method='migrad')

~/phd/git/pyerrors_github/pyerrors/fits.py in least_squares(x, y, func, priors, silent, **kwargs)
    118         return _prior_fit(x, y, func, priors, silent=silent, **kwargs)
    119     else:
--> 120         return _standard_fit(x, y, func, silent=silent, **kwargs)
    121 
    122 

~/phd/git/pyerrors_github/pyerrors/fits.py in _standard_fit(x, y, func, silent, **kwargs)
    501         chisquare = fit_result.fun
    502 
--> 503         output.iterations = fit_result.nit
    504     else:
    505         output.method = 'Levenberg-Marquardt'

/opt/anaconda3/lib/python3.6/site-packages/scipy/optimize/optimize.py in __getattr__(self, name)
    122             return self[name]
    123         except KeyError:
--> 124             raise AttributeError(name)
    125 
    126     __setattr__ = dict.__setitem__

AttributeError: nit

using python3.6 and migrad 2.8.4 (and 2.8.3)

[fjosw]

In the back of my head I was aware that the migration to iminuit 2.0 broke this routine. The question is whether we want to keep the option for several solver in the least_squares routine or just support the scipy one.

[s-kuberski]

I like the possibility to switch to migrad, just to see if everything is stable.