Add additional checks to InvalidTargetDataCheck to handle invalid target data types #929
Conversation
Codecov Report
@@ Coverage Diff @@
## main #929 +/- ##
=======================================
Coverage 99.85% 99.86%
=======================================
Files 170 170
Lines 8565 8593 +28
=======================================
+ Hits 8553 8581 +28
Misses 12 12
Continue to review full report at Codecov.
|
| return [DataCheckError("{} row(s) ({}%) of target values are null".format(null_rows.sum(), null_rows.mean() * 100), self.name)] | ||
| if null_rows.any(): | ||
| messages.append(DataCheckError("{} row(s) ({}%) of target values are null".format(null_rows.sum(), null_rows.mean() * 100), self.name)) | ||
| numerics = ['int16', 'int32', 'int64', 'float16', 'float32', 'float64', 'bool'] |
There was a problem hiding this comment.
We're also defining numerics in PR #917. Might be best to define once in a common location and then import it wherever its needed.
There was a problem hiding this comment.
Agreed! And same with having supported_data_types somewhere common.
| value_counts = y.value_counts() | ||
| if len(value_counts) == 2 and y.dtype in numerics: | ||
| unique_values = value_counts.index.tolist() | ||
| if set(unique_values) != set([0, 1]): |
There was a problem hiding this comment.
What happens if unique_values has float dtype? Does this still work properly?
There was a problem hiding this comment.
Yup! But added test_invalid_target_data_check_numeric_binary_classification_valid_float so we don't have to wonder 😄
There was a problem hiding this comment.
@angela97lin Good to merge but please remember to consolidate the definition of numerics in #932 !
Closes #916