Adds __repr__ for our custom classes

[billbrod]

This adds a __repr__ method for our custom classes: basis, glm, observation models, and regularizers. The logic for almost all of these are included in the new nmo.utils.format_repr function, summarized as:

- The returned string includes the output of ``obj.get_params(deep=False)`` if value
  (that is, if it's not False, None, empty, etc.).

- Any key in ``exclude_keys`` are excluded.

- If the key is in ``use_name_keys``, we use ``value.__name__``. Else we use
  ``value.__repr__``

TransformerBasis and CompositeBasis are different:

• TransformerBasis is Transformer(self.basis)
• CompositeBasis is

AdditiveBasis(
    basis1=self.basis1,
    basis2=self.basis2,
)

and same for Multiplicative, with composites of composites using additional tab indenting

[BalzaniEdoardo]

thanks for the first pass at this!

I added a couple of comments. I would include a max depth or something similar for additive bases in the repr.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.40%. Comparing base (a510ef3) to head (67da7b4).
Report is 110 commits behind head on development.

Additional details and impacted files

@@               Coverage Diff               @@
##           development     #286      +/-   ##
===============================================
+ Coverage        96.13%   97.40%   +1.26%     
===============================================
  Files               34       35       +1     
  Lines             2642     2850     +208     
===============================================
+ Hits              2540     2776     +236     
+ Misses             102       74      -28     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[BalzaniEdoardo]

Added Features

• Slightly changed the format_repr utility function:
  • Ignore parameters that have the attribute shape (intended for arrays).
  • Do not ignore parameters with value False and 0, while ignore other falsey values like None or "".
• Added basic repr tests.
• Doctests for basis.

Note: I did not add doctests for GLM repr because the output is way too long, but I have included actual tests.

[billbrod]

Should probably add a doctest for at least the GLM (if not also regularizer and observation models?)

There are other notes scattered throughout

[billbrod]

Most recent change adds label to the repr. This only affects basis right now (others don't have a label param). Questions about that:

• We don't handle label="" (which user would have to do themselves) in a special way, so you'd get back e.g., (RaisedCosineLinearEval, width=2.0, ...). (With label=None and if the user doesn't set it directly, you'd get RaisedCosineLinearEval(width=2.0, ...))
• I didn't add the label behavior to any doctests, because I didn't see it being used in any existing doctests. Should I?
• User can't set label for Additive and Multiplicative basis, so I don't handle it specifically there. But they do have a more informative label -- should I make use of it?
• With this, I don't think we're using exclude_keys arg anywhere (it's being used in for the GLM, but as I noted in my comments, I think that can be removed) -- should we keep it in case we need it later or remove it?

[sjvenditto]

I've added suggestions for the composite basis repr to have a maximum print length based on the terminal size. The line divisor might be too big (i.e. the print underfills the terminal), but here's an example of a line limited repr in my terminal: