[SciPy-Dev] On the leastsq/curve_fit method

[Martin Teichmann]

Hello everybody,

I just saw that there is some discussion about leastsq, and
guessed that this is a good time to add some ideas.

I have been working on an all-python fitter, which is just
the minpack fitter translated to python. This is why I was
bugging yall for the qr_multiply stuff, for those who remember.

Having the fitter in python is quite useful, because this
calling python from FORTRAN sometimes was ugly: most
importantly because exceptions often didn't find their way
through FORTRAN.

I added some new stuff as well: if the fitted function realizes
that it's parameters are running away, it may raise an
InvalidParameter exception, and the fitting algorithm will
choose a smaller step size on this parameter.

You can find the code at https://github.com/scipy/scipy/pull/88

Greetings

Martin