Join failed to get groupBy module on different file structure

[MickJermsurawong]

Summary

I made a quick repro and curious on your input. I think when people define chronon features, it's expected they follow the structure, but it can live within some directories.. In that way, i think allowing flexible top module should be allowed. And we should be able to refer to chronon features object like other piece of python code. In this PR,

• i'm adding a new unit test, simply importing the chronon features in unit test, and pytest api/py/test/test_sample.py will fail

api/py/test/sample/joins/quickstart/training_set.py:20: in <module>
    from group_bys.quickstart.purchases import v1 as purchases_v1
E   ModuleNotFoundError: No module named 'group_bys'

• i have to name change namespace import from group_bys.quickstart.returns to sample.group_bys.quickstart.returns to make import work, now it fails couldn't find the module, although it's defined in a different file.

ERROR project/chronon_features/chronon_features/tests/joins/test_quickstart_training_set.py - 
ValueError: [GroupBy] Must specify a group_by name if group_by is not defined in separate file. You may pass it in via GroupBy.name.

• Lastly, i updated the check and test pass

Why / Goal

Test Plan

• Added Unit Tests
• Covered by existing CI
• Integration tested

Checklist

• Documentation update

Reviewers

[nikhilsimha]

ah, the fix here is to set python path to your top level chronon configuration directory.

if vscode recognizes the path, compile.py should too.

[MickJermsurawong]

thanks nikhil, yup appending python path works.. for now i have to add add pythonpath in unit test, wrapping for compile.py, and add ignore on mypy. I think it would be great if we can treat these definitions like native python resources using the standard python module references

[nikhilsimha]

They should work like standard python references actually - it is just that the python env needs to be setup to recognize chronon on the module. Or maybe I am misunderstanding, is it a wrong assumption for the framework to expect the top level chronon directory (sample in this case) to be on python path?