Deprecate cudf.isclose #17351
Deprecate cudf.isclose #17351
Conversation
github-actions
bot
added
libcudf
| @@ -5,17 +5,17 @@ | |||
| "args": { | |||
| "CUDA": "11.8", | |||
| "PYTHON_PACKAGE_MANAGER": "conda", | |||
| "BASE": "rapidsai/devcontainers:24.12-cpp-cuda11.8-mambaforge-ubuntu22.04" | |||
| "BASE": "rapidsai/devcontainers:25.02-cpp-cuda11.8-mambaforge-ubuntu22.04" | |||
There was a problem hiding this comment.
This PR targets 24.12. Why have all the dependencies been changed to 25.02?
There was a problem hiding this comment.
The PR's title, name, description, and actual changes do not make sense to me.
Recommend closing this.
There was a problem hiding this comment.
This appears to be an attempt at #13593, an issue I filed a while back. @the-gates-of-Zion I can help work with you on this, if you are interested. This should target branch-25.02, so I will retarget the PR.
| @@ -5322,6 +5322,29 @@ def isclose(a, b, rtol=1e-05, atol=1e-08, equal_nan=False): | |||
| 5 False | |||
| dtype: bool | |||
| """ | |||
| warnings.warn( | |||
| "`cudf.close` is deprecated and will be removed in a future version of cudf. " | |||
There was a problem hiding this comment.
| "`cudf.close` is deprecated and will be removed in a future version of cudf. " | |
| "`cudf.isclose` is deprecated and will be removed in a future version of cudf. " |
| ''' | ||
| import cupy as cp | ||
| import pandas as pd | ||
| from cudf.core.column import ( | ||
| as_column, | ||
| ) | ||
|
|
||
| a = pd.array([1.0, 2.0, None]) | ||
| b = pd.array([1.0, 2.1, None]) | ||
|
|
||
| a_col = as_column(a) | ||
| a_array = cupy.asarray(a_col.data_array_view(mode="read")) | ||
|
|
||
| b_col = as_column(b) | ||
| b_array = cupy.asarray(b_col.data_array_view(mode="read")) | ||
|
|
||
| result = cp.isclose(a, b, equal_nan=True) | ||
| print(result) # Output: [ True False True] | ||
| ''', |
There was a problem hiding this comment.
We noted in #13593: "As part of the deprecation, we decided to add a warning message indicating how to support nulls while using cupy.isclose." However, I am not sure if this is the right code snippet.
The code we want to give to the user as a replacement for this deprecated code path is essentially the current implementation in this method. Basically it has two steps:
- Call
cupy.iscloseto see if the values are "close" to each other. This does not account for null values in the input. - Compare the null masks of each input. If
equal_nanisFalse, we mark all null values asFalse("not close") in the result. Ifequal_nanisTrue, we set the values where one column's input is null asFalseand the values where both columns' input are null asTrue.
I would suggest we add this snippet to the page "Working with missing data": https://docs.rapids.ai/api/cudf/stable/user_guide/missing-data/
Then we can link to this snippet in the docs from the deprecated method's docstring and in the deprecation warning.
Does that make sense? Let me know if you have questions.
Description
add warnings.warn("
cudf.closeis deprecated and will be removed in a future version of cudf. ", ''' ... ''')(from 25.02)
Checklist