large mnl simulation fails if choosers is a list

[mxndrwgrdnr]

According to the definition of large_mnl, both choosers and out_choosers can be either string type or list of string type, seen here and here, respectively. This all works fine until you get to simulation which uses orca.get_table() function on choosers/out_choosers to perform the column update and save results, seen here. I'm pretty sure this could be fixed by replacing orca.get_table() with a self._get_df() call like is done elsewhere in this model class, but perhaps it would be better to create an additional class attribute model.out_table to refer specifically to the output table that will store the results, and which can't be a list of tables, and would therefore be distinct from either model.choosers or model.out_choosers. Any thoughts?

[mxndrwgrdnr]

Actually I think the second idea, using a new argument out_table is the only one that will work because you cannot call update_col_from_series on the output of _get_df().

[smmaurer]

Oh, good catch. I think the reasoning we've used in other places is that the first table in a list needs to be the "primary" table, i.e. the one whose rows represent the unit of analysis -- otherwise it's impossible to know how to merge the tables together properly. And in this case, we can assume/require that the output variable goes into the first table.

See here in the OLS docstrings for example. So i guess we need to replace MNL#L524-L527 with something like TemplateStep#L290-L308 that i wrote for OLS..

[mxndrwgrdnr]

Ah, yes, that looks like a more elegant solution. I'll submit the PR.

[mxndrwgrdnr]

Fixed in #69.