fix benchmark failure with numpy 1.20+ #39795
Conversation
| @@ -135,7 +135,7 @@ def time_isin_long_series_long_values_floats(self): | |||
|
|
|||
| class IsInLongSeriesLookUpDominates: | |||
| params = [ | |||
| ["int64", "int32", "float64", "float32", "object", "Int64", "Float64"], | |||
| ["int64", "int32", "float64", "float32", "object"], | |||
There was a problem hiding this comment.
Can you instead raise a NotImplementedError in the setup below if dtype in ("Int64", "Float64") and np.__version >+ "1.20.0" ?
That keeps the benchmark running for if you don't have numpy 1.20 installed (although I don't know how often the environment gets updated on our benchmark server)
There was a problem hiding this comment.
Thanks @jorisvandenbossche . done.
There was a problem hiding this comment.
Thanks!
Further follow-up question: 1) is there actually an issue about this failure? 2) can you add a link to that issue as a TODO comment where raising the NotImplementedError?
There was a problem hiding this comment.
is there actually an issue about this failure?
can you add a link to that issue as a
TODOcomment where raising the NotImplementedError?
as a follow-up
|
thanks @simonjayhawkins |
|
@meeseeksdev backport 1.2.x |
|
changed milestone, backport not needed. see #39842 (comment) |
xref #38379 (review)
This will be blocking using numpy 1.20 on Github actions Checks, xref #36092 (comment)