Added missing attributes to other fit methods

[omdv]

Greetings!

I started using LMFIT for my own project, where I switch between different fitting methods often and compare goodness of fit using reduced chisquared and Rsquared. I have discovered that all methods except least squares don't have defined redchi and residuals attributes. It was even pointed out on stackoverflow: http://stackoverflow.com/questions/17465701/attribute-error-from-minimizer-object-returned-from-scipy-optimize-minimize-fu

Hence the following changes were made to lbfgsb(), fmin() and scalar_minimize():

• Residual attribute added to the remaining methods. It is based on calling __residual() with optimized parameters
• Chisqr calculation is based on residuals
• Redchi added
• leastsq method is not changed

I tested this version with my project and haven't seen any issues. It also passed all available tests.
Please let me know if you have any questions or concerns.

Cheers

[newville]

Thanks! I agree having chisqr and redchi would be nice for scalar minimizers. I'll merge this, but then alter it to use (as I think one of your earlier versions did) Minimizer.penalty() -- the scalar minimizing functions allow the objective function to return a single number. For your use case (and, I expect the most common), this would be exactly "chi-square". But it doesn't have to be that. Still, providing a redchi as if the users objective function returned either chi-square or an ndarray to minimize in the the least-squares sense is perfectly reasonable.

[EnboYang]

Is there add "R square" to fit report?
https://en.wikipedia.org/wiki/Coefficient_of_determination

[newville]

@EnboYang This PR was merged almost a year ago. If you are requesting a further change to this code, you're too late. If you're asking a question about a new feature, you're in the wrong place.

Please don't mix questions with closed PRs -- the conversations are too hard to follow and find. Use the lmfit-py mailing list https://groups.google.com/forum/#!forum/lmfit-py