Added fast dispersion fitter

[caseyflex]

Features:

• Passivity enforcement and lossless materials via lower and upper bounds on loss
• Weighting real / imaginary part
• Fitting of eps_inf

Known limitations:

• Running with eps_inf=None can give worse results than running with fixed eps_inf
• Works quite well on every material in library except the following two:
  • Pd-JohnsonChristy1972, Ni-JohnsonChristy1972

[dbochkov-flexcompute]

some random feedback, not sure if it is already addressed: I happened to play with different fittings recently, and it seemed like fit_fast ignores wvl_range parameter, i.e. it always tries to fit all provided data?

[caseyflex]

some random feedback, not sure if it is already addressed: I happened to play with different fittings recently, and it seemed like fit_fast ignores wvl_range parameter, i.e. it always tries to fit all provided data?

thanks, should be fixed now

[weiliangjin2021]

Looks quite good! It now works well with several metals in my tests. Since we start to support gain medium recently, to fit a gain medium, is it by either setting loss_bounds=(-np.inf, np.inf) or passivity_num_iters=0? As for the latter, it is currently a PositiveInt, so might be good to change it to an int?

[caseyflex]

Looks quite good! It now works well with several metals in my tests. Since we start to support gain medium recently, to fit a gain medium, is it by either setting loss_bounds=(-np.inf, np.inf) or passivity_num_iters=0? As for the latter, it is currently a PositiveInt, so might be good to change it to an int?

Thanks for the comments. It's by setting loss_bounds. I improved the docstrings to make this more clear.

[tylerflex]

Looks good to me, I don't understand many of the internals so most of my comments are pretty surface level. More generally, are we considering adding more fitters in the future? If so, it would probably be nice to have a more unified interface in terms of parameters, the general .fit() method (what is displayed, how progress bars work), and the testing. Thanks Casey!

[tylerflex]

Also, what I've been trying out is marking PRs as "draft" before they are ready for review and then marking "ready for review" when it is ready to review and merge.

[lucas-flexcompute]

I don't have much to add from my previous comments or Tyler's points. One thing I noticed is that max_num_poles and min_num_poles seem very similar in purpose, but are accessible in different parts of the API. Is there any deeper reason for this choice (maybe compatibility with other fitters)?

Other than that, I believe we talked about making this fitter the easy default for the user. Is this change in API supposed to be included in this PR? In that case, it should probably be followed by changes in the docs.

[caseyflex]

Looks good to me, I don't understand many of the internals so most of my comments are pretty surface level. More generally, are we considering adding more fitters in the future? If so, it would probably be nice to have a more unified interface in terms of parameters, the general .fit() method (what is displayed, how progress bars work), and the testing. Thanks Casey!

Thanks for the comments. I don't think we are considering adding more fitters in the future. Just keeping the stable fitter as a backup.

[caseyflex]

I don't have much to add from my previous comments or Tyler's points. One thing I noticed is that max_num_poles and min_num_poles seem very similar in purpose, but are accessible in different parts of the API. Is there any deeper reason for this choice (maybe compatibility with other fitters)?

Other than that, I believe we talked about making this fitter the easy default for the user. Is this change in API supposed to be included in this PR? In that case, it should probably be followed by changes in the docs.

Thanks, changed this so they are both optional arguments now. As discussed, I will change the fitter notebook to showcase this one as the default.

[tylerflex]

@weiliangjin2021 @lucas-flexcompute if you both approve of the PR can you formally approve it through GitHub and then I'll merge it. If you want changes, please request them. thanks!