[0.1.dev17] latest updates to segmented MNL

[mxndrwgrdnr]

this PR addresses:

• Segmented MNL fails to read list of tables for alts or choosers #58
• latest changes to choicemodels library (specifically this PR)
• adds a couple new tests to reflect the changes

[smmaurer]

This looks good! Some comments:

1. Should add docstrings for LargeMultinomialLogitStep.run() explaining that chooser_batch_size will be passed through to iterative_lottery_choices(), and is a temporary solution that will eventually be replaced by a class parameter/setting (i can take care of it with issue Settings objects for groups of common parameters #54).

2. In the same place, instead of adding an interaction_terms argument, would you want to just accept a custom MergedChoiceTable like is done higher up in the fit() method? This is more flexible, but i guess might be more work for the user. Either one would be temporary -- issue Settings objects for groups of common parameters #54 will include a solution for orca to manage interaction terms.

3. Can you bump the version number in setup.py and the top-level __init__.py? The current master branch is 0.1.dev16, so you can make this dev17. I'll wait to merge the other open PR, and can change it to dev18.

The tests look good and all pass on my machine.

I like your temporary solution for printing the segment name. Ok if i wait to implement the full solution from issue #59? We can leave it open so i don't forget.

[mxndrwgrdnr]

Thanks for the notes Sam. Here are my answers:

1. Done
2. I agree that would be more consistent with how the mct is specified in the fit() method, but a) I think that might defeat the purpose of the chooser batching since the idea there is to delay the creation of the mct's until the batches have been chunked out to avoid having to hold a giant table in memory; and b) it would require more changes downstream to choicemodels code since currently iterative_lottery_choices() requires alts and obs to be passed in separately. It would be way more straightforward to implement what you suggest if we didn't care about the constrained choice scenario, but I don't see an easy way to do it otherwise. Open to any ideas you might have here, though.
3. Done

Also, yes, sorry, I actually didn't mean to commit the segment name printing change to the code so I'm happy to ignore it.

[smmaurer]

2. Hadn't thought of that. Let's stick with your approach!

[smmaurer]

And thanks for the changes! Looks great to merge.