BUG: Fix np.quantile([Fraction(2,1)], 0.5) #24711
Merged
+6
−1
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
The debug test failure seems to be an Ubuntu problem, the requested file is missing from the upstream directory. I suspect it is due to a micro upgrade in the Ubuntu version. |
seberg
reviewed
Sep 19, 2023
numpy/lib/_function_base_impl.py
Outdated
| @@ -4579,7 +4579,8 @@ def _lerp(a, b, t, out=None): | |||
| diff_b_a = subtract(b, a) | |||
| # asanyarray is a stop-gap until gh-13105 | |||
| lerp_interpolation = asanyarray(add(a, diff_b_a * t, out=out)) | |||
| subtract(b, diff_b_a * (1 - t), out=lerp_interpolation, where=t >= 0.5) | |||
| subtract(b, diff_b_a * (1 - t), out=lerp_interpolation, where=t >= 0.5, | |||
| casting='unsafe') | |||
There was a problem hiding this comment.
This may sound weird, but we should probably also add dtype=type(lerp_interpolation.dtype).
If we have unsafe casting, this means that at least to some degree we ignore the precision of the weights of the quantiles for the output precision (promotion). Can you give me a hint in how things are currently? Does the same happen if we use np.longdouble for the quantile level (q)?
(Sorry, but I am a bit surprised how many subtleties there still seem to be along promotion in quantiles :(, and hoping we can clarify things a bit for me and everyone else.)
There was a problem hiding this comment.
@seberg I tested with
import numpy as np
print(np,np.__version__)
x = np.linspace(.1, .7, 5, dtype=np.longdouble)
for q in [0, 1, 0., 1., .5]:
r = np.quantile(x, q)
print(f'{q} ({type(q)}) -> {r} ({type(r)})')
r = np.quantile(x, np.longdouble(q))
print(f'{q} ({type(q)}) -> {r} ({type(r)})')
print('----')
x = np.array([0, 1])
for q in [0, 1, 0., 1., .5]:
r = np.quantile(x, q)
print(f'{q} ({type(q)}) -> {r} ({type(r)})')
r = np.quantile(x, np.longdouble(q))
print(f'{q} ({type(q)}) -> {r} ({type(r)})')
Output on 1.26:
<module 'numpy' from '/home/eendebakpt/py310/lib/python3.10/site-packages/numpy/__init__.py'> 1.26.0
0 (<class 'int'>) -> 0.1 (<class 'numpy.longdouble'>)
0 (<class 'int'>) -> 0.1 (<class 'numpy.longdouble'>)
1 (<class 'int'>) -> 0.7 (<class 'numpy.longdouble'>)
1 (<class 'int'>) -> 0.7 (<class 'numpy.longdouble'>)
0.0 (<class 'float'>) -> 0.1 (<class 'numpy.longdouble'>)
0.0 (<class 'float'>) -> 0.1 (<class 'numpy.longdouble'>)
1.0 (<class 'float'>) -> 0.7 (<class 'numpy.longdouble'>)
1.0 (<class 'float'>) -> 0.7 (<class 'numpy.longdouble'>)
0.5 (<class 'float'>) -> 0.4 (<class 'numpy.longdouble'>)
0.5 (<class 'float'>) -> 0.4 (<class 'numpy.longdouble'>)
----
0 (<class 'int'>) -> 0 (<class 'numpy.int64'>)
0 (<class 'int'>) -> 0.0 (<class 'numpy.longdouble'>)
1 (<class 'int'>) -> 1 (<class 'numpy.int64'>)
1 (<class 'int'>) -> 1.0 (<class 'numpy.longdouble'>)
0.0 (<class 'float'>) -> 0.0 (<class 'numpy.float64'>)
0.0 (<class 'float'>) -> 0.0 (<class 'numpy.longdouble'>)
1.0 (<class 'float'>) -> 1.0 (<class 'numpy.float64'>)
1.0 (<class 'float'>) -> 1.0 (<class 'numpy.longdouble'>)
0.5 (<class 'float'>) -> 0.5 (<class 'numpy.float64'>)
0.5 (<class 'float'>) -> 0.5 (<class 'numpy.longdouble'>)
Output with this PR:
<module 'numpy' from '/mnt/data/numpy/numpy/__init__.py'> 2.0.0.dev0+git20231107.a555e7d
0 (<class 'int'>) -> 0.1 (<class 'numpy.longdouble'>)
0 (<class 'int'>) -> 0.1 (<class 'numpy.longdouble'>)
1 (<class 'int'>) -> 0.7 (<class 'numpy.longdouble'>)
1 (<class 'int'>) -> 0.7 (<class 'numpy.longdouble'>)
0.0 (<class 'float'>) -> 0.1 (<class 'numpy.longdouble'>)
0.0 (<class 'float'>) -> 0.1 (<class 'numpy.longdouble'>)
1.0 (<class 'float'>) -> 0.7 (<class 'numpy.longdouble'>)
1.0 (<class 'float'>) -> 0.7 (<class 'numpy.longdouble'>)
0.5 (<class 'float'>) -> 0.4 (<class 'numpy.longdouble'>)
0.5 (<class 'float'>) -> 0.4 (<class 'numpy.longdouble'>)
----
0 (<class 'int'>) -> 0 (<class 'numpy.int64'>)
0 (<class 'int'>) -> 0.0 (<class 'numpy.longdouble'>)
1 (<class 'int'>) -> 1 (<class 'numpy.int64'>)
1 (<class 'int'>) -> 1.0 (<class 'numpy.longdouble'>)
0.0 (<class 'float'>) -> 0.0 (<class 'numpy.float64'>)
0.0 (<class 'float'>) -> 0.0 (<class 'numpy.longdouble'>)
1.0 (<class 'float'>) -> 1.0 (<class 'numpy.float64'>)
1.0 (<class 'float'>) -> 1.0 (<class 'numpy.longdouble'>)
0.5 (<class 'float'>) -> 0.5 (<class 'numpy.float64'>)
0.5 (<class 'float'>) -> 0.5 (<class 'numpy.longdouble'>)
If you want I can add tests for this case as well.
0f8b04d to
577d8d4
Compare
|
@seberg ping. |
|
Thanks @eendebakpt, sorry for forgetting to look over this again. I think it should be fine to do this cast and it seems like you gave it a spin with that test. So let's give it a shot. |
charris
pushed a commit
to charris/numpy
that referenced
this pull request
Feb 3, 2024
* BUG: Fix np.quantile([Fraction(2,1)], 0.5) * address review comments * pass type instead of dtype instance
charris
added a commit
that referenced
this pull request
Feb 3, 2024
BUG: Fix np.quantile([Fraction(2,1)], 0.5) (#24711)
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
On main:
In this PR we update the
np.quantileto handle the casenp.quantile(arr, 0.5)as well.Also see #24592