Feature/sbachmei/test column specific noising #131
Conversation
stevebachmeier
requested review from
albrja,
hussain-jafari,
mattkappel,
ramittal and
rmudambi
as code owners
…/sbachmei/test-column-specific-noising
…/sbachmei/test-column-specific-noising
| } | ||
|
|
||
| # Set row-level noise to zero since we're just testing columns here. | ||
| config[dataset.name][Keys.ROW_NOISE] = { |
There was a problem hiding this comment.
If row noise is not present, don't we have guarantees that the indexes will match in the case where we're using the sample data?
Why aren't we testing row noise in this test? Do we have another test that is testing that row noise is applied?
There was a problem hiding this comment.
Good point - probably? I added this because the targeted omission stuff was failing. Haven't thought through yet a better way to do it with that.
As for row noise - that just seems like a different type of test and out of scope for this ticket. Maybe I'm wrong.
There was a problem hiding this comment.
After discussing w/ Rajan, we will go ahead with this approach and then immediately follow it up w/ a bugfix task
| for col in noised_data.columns: | ||
| expected_dtype = [c.dtype_name for c in COLUMNS if c.name == col][0] | ||
| # Check each column. We set the index for each dataset to be unique | ||
| # identifiers b/c the original index gets reset after noising. Note that |
There was a problem hiding this comment.
Where does the index get reset? At the dataset level? Not in the noise function for columns.
There was a problem hiding this comment.
Title: Test expected columns are noised
Description
Currently for integration testing we are only really ensuring that the noised
dataset is at all different than the raw dataset. This PR checks
column-by-column that the expected cols are different and those that
are output but not noised are the same.
Testing
tests pass