Improve visual_test.py #3732
Improve visual_test.py #3732
Conversation
Do not run the script when running pytest on an inplace build. Clean-up and speed-up (10x) calculations of the donuts volume. Many style fixes, passing flake8.
| @@ -3,70 +3,118 @@ | |||
| data. | |||
| """ | |||
|
|
|||
| from __future__ import print_function | |||
There was a problem hiding this comment.
We don't support Py2.7 on master so this is unnecessary.
There was a problem hiding this comment.
I added it for backport to 0.14.x
There was a problem hiding this comment.
My policy generally has been to leave any backport-required code for the backport, but to each their own. ;)
| x1 *= x1 | ||
| x2 = x + y2 + zc | ||
| x2 *= x2 | ||
| vol[iz, iy, ix] = ((x1 - d * (x + y1)) * (x2 - d * (z + y2))) |
There was a problem hiding this comment.
This looks eminently vectorisable...?
There was a problem hiding this comment.
I agree that it'll use more memory, but I disagree that it'll be less readable. It would presumably read like an actual formula, which right now I can't for the life of me figure out based on this code. But on the other hand this was even more true in the previous version of the code! So, as mentioned before, this is fine. =)
There was a problem hiding this comment.
Can we at least agree that constructs such as:
z = iz * a + b
z *= zshould be replaced by
z = (iz * a + b) ** 2There are many of them in here, and github's suggestion feature doesn't see appropriate.
There was a problem hiding this comment.
I guess not, it makes things slower. 40 ms as written, and 47 ms with the ( ) ** 2 suggestion.
Using fully vectorized code, it goes down to 2ms
def donuts():
"""Return volume of two donuts based on a formula by Thomas Lewiner."""
n = 48
a = 2.5 / n * 8.0
b = -1.25 * 8.0
c = 16.0 - 1.85 * 1.85
d = 64.0
i = np.arange(n, dtype=int)
ia_plus_b = i * a + b
ia_plus_b_square = ia_plus_b ** 2
z = ia_plus_b_square[:, np.newaxis, np.newaxis]
zc = z + c
y1 = ((ia_plus_b - 2) ** 2)[np.newaxis, :, np.newaxis]
y2 = ((ia_plus_b + 2) ** 2)[np.newaxis, :, np.newaxis]
x = ia_plus_b_square[np.newaxis, np.newaxis, :]
x1 = (ia_plus_b_square + y1 + zc) ** 2
x2 = (ia_plus_b_square + y2 + zc) ** 2
return ((x1 - d * (x + y1)) * (x2 - d * (z + y2))) + 1025
There was a problem hiding this comment.
I guess not, it makes things slower. 40 ms as written, and 47 ms with the
( ) ** 2suggestion.
Well, in this case, readability trumps speed, or we wouldn't be doing such a loop in Python.
But I agree that the vectorised version is better.
|
Hello @cgohlke! Thanks for updating the PR. Cheers ! There are no PEP8 issues in this Pull Request. 🍻 Comment last updated on February 10, 2019 at 19:22 Hours UTC |
|
I don't mind pep8 so much for whiltespace lines, most of my editors remove them automatically. spyder wasn't configured correctly... |
|
@cgohlke I think you are right that writing a for loop isn't a terrible idea, but this would be a perfect place where a project like pythran might be useful.
Whats the review process say now @jni. I can't merge this myself can I? Do we have to wait for an other core-dev? |
Great questions all! In fact we don't have a guideline for this. In the past my internal guide has been "merge ok if changes are minor (typos, comments, variable naming), wait if changes are significant (major code restructuring, logic changes, and definitely for API changes)". We should probably add something to the core dev guide about this. In this specific case, the changes are significant, but it is a minor file that is rarely run, so I'd be ok with merging if @cgohlke is ok with the changes you made. (Personally I like them. And I've verified locally that it works.) @cgohlke? |
|
Yeah. Sorry about just pushing to your branch. I figured you knew how to revise to remove the changes ;) |
Description
Do not run the script when running pytest on an inplace build.
Clean-up and speed-up (10x) calculations of the donuts volume.
Many style fixes, passing flake8.
Checklist
./doc/examples(new features only)./benchmarks, if your changes aren't covered by anexisting benchmark
For reviewers
later.
__init__.py.doc/release/release_dev.rst.@meeseeksdev backport to v0.14.x