Move list_model_families to avoid circular dependencies #959
Conversation
dsherry
added
enhancement
Codecov Report
@@ Coverage Diff @@
## main #959 +/- ##
==========================================
- Coverage 99.87% 99.87% -0.01%
==========================================
Files 171 171
Lines 8783 8780 -3
==========================================
- Hits 8772 8769 -3
Misses 11 11
Continue to review full report at Codecov.
|
dsherry
force-pushed
the
ds_911_circular_import
branch
from
457d095
to
50661ac
Compare
| @@ -102,7 +102,7 @@ def __init__(self, | |||
| allowed_model_families to be ignored. | |||
|
|
|||
| allowed_model_families (list(str, ModelFamily)): The model families to search. The default of None searches over all | |||
| model families. Run evalml.list_model_families("binary") to see options. Change `binary` | |||
| model families. Run evalml.pipelines.components.utils.allowed_model_families("binary") to see options. Change `binary` | |||
There was a problem hiding this comment.
I thought about adding this to the user guide but I like what's currently there.
dsherry
force-pushed
the
ds_911_circular_import
branch
from
50661ac
to
9317ce7
Compare
|
Oof, codecov is complaining because I made a docstring change to a file with near-100% coverage. <0.01% project coverage change introduced by this PR. I plan to ignore that and merge. Perhaps we should lower our codecov threshold... |
There was a problem hiding this comment.
@dsherry Looks good to me! I think its fine to merge without the project codecov test passing in this instance.
Notes here
Summary: we ran into a circular dependency issue in this code. We realized that since
list_model_familiesis currently defined as an attribute of estimators, we should move that code into the components namespace, and that doing so would avoid issues with importing other utilities from the components namespace.Building on work from #898 #911 #934