Adding IsYearEnd and IsYearStart primitives #2124
Conversation
Codecov Report
@@ Coverage Diff @@
## main #2124 +/- ##
=======================================
Coverage 99.21% 99.21%
=======================================
Files 143 143
Lines 16837 16869 +32
=======================================
+ Hits 16705 16737 +32
Misses 132 132
Continue to review full report at Codecov.
|
|
|
||
| def test_is_year_end(): | ||
| is_year_end = IsYearEnd() | ||
| dates = pd.Series([datetime(2020, 12, 31), datetime(2020, 1, 1)]) |
There was a problem hiding this comment.
What happens if there is a NaN?
| >>> from datetime import datetime | ||
| >>> dates = [datetime(2019, 12, 31), | ||
| ... datetime(2019, 1, 1), | ||
| ... datetime(2019, 11, 30)] |
There was a problem hiding this comment.
Can we include an example with NaN?
There was a problem hiding this comment.
We should make sure we did this with the previous Datetime primitives we merged in
There was a problem hiding this comment.
General question regarding the behavior of primitives with NaN:
If I use the Month primitive on the following data:
pd.Series([datetime(2020, 3, 3), np.nan])
I get [3.0, nan]
If I pass pd.Series([np.nan]) in to the Month primitive, I get an error.
I get analogous behavior for Year and the like. I just wanted to confirm that this is indeed the correct expected behavior before writing the tests.
There was a problem hiding this comment.
I think you may have found a bug :)
There was a problem hiding this comment.
You should always expect the series dtype for Datetime columns to be datetime64[ns] based on the behavior of Woodwork, which enforces that dtype for the series.
If you don't otherwise specify the dtype a series of pd.Series([np.nan]) will have a dtype of float64 and you can't perform datetime operations on a float64 column.
I haven't tested this out, but I think if you convert pd.Series([np.nan]) to a datetime column first, the Month primitive will no longer error. If you still get an error at that point, then I agree with @dvreed77 there is a bug we need to fix.
There was a problem hiding this comment.
Good call. I confirmed this doesn't error
s = pd.Series([np.nan]).astype('datetime64[ns]')
ft.primitives.Month()(s)…eryx/featuretools into more-year-and-day-primitives
| Examples: | ||
| >>> from datetime import datetime | ||
| >>> dates = [datetime(2019, 12, 31), | ||
| ... datetime(2019, 1, 1), |
There was a problem hiding this comment.
@sbadithe can you put a NaN example in the doc string?
Adding the following primitives: IsYearEnd, IsYearStart
Fixes #2061
Fixes #2060