Rom Operator Inference Python3 Versions Save

Large restructuring of the code and many API changes. Migrating rom_operator_inference to opinf on PyPI after this release.

Some under-the-hood improvements to lstsq solver classes. See #18. API changes:

• OpInf ROM fit() methods no longer take additional arguments for least-squares solvers because there are no longer additional arguments for any of the solvers.
• All lstsq solver classes have signatures fit(A, B), some were fit(A,b) previously.
• All lstsq solver classes now have cond(), regcond(), misfit(), and residual() methods for post analysis.

Mostly internal fixes, with a few API changes:

• utils.expand_Hc() -> utils.expand_H()
• utils.expand_Gc() -> utils.expand_G()
• utils.lstsq_reg() -> lstsq.solve()
• new submodule lstsq groups tools for solving the least-squares problems for Operator Inference

Mostly documentation updates, plus a few bug fixes in the least squares wrapper and throughout.

This release makes some API changes to the pre module:

• pre.mean_shift() is now pre.shift()
• new function pre.scale()
• pre.significant_svdvals() is now pre.svdval_decay()
• pre.energy_capture() is now pre.cumulative_energy()

The old names are still available but will issue a DeprecationWarning.

This version also makes various minor improvements in the code and documentation.

This release fixes some bugs with model i/o in ROMclass.save_model() and roi.load_model(). It also adds the option to set Vr=None in inferred class fit() methods to save on memory if the pre-projected states are available for training.

See https://pypi.org/project/rom-operator-inference/.