Make all zips explicitly strict or non-strict #850

Armavica · 2024-06-24T14:27:55Z

Description

First commit: adding a strict=True argument to all zips when it doesn't produce mistakes in the test suite (464 of them), and strict=False to the others (28 of them)
Second commit: enable ruff rule requiring and explicit strict argument to all zips
Rest of the commits: transform the non-strict zips into strict zips (18 of them for now)

There remains 10 non-strict zips that I find difficult to understand.

Related Issue

Closes #
Related to Make zip strict #840

Checklist

Checked that the pre-commit linting/style checks pass
Included tests that prove the fix is effective or that the new feature works
Added necessary documentation (docstrings and/or example notebooks)
If you are a pro: each commit corresponds to a relevant logical change

Type of change

ricardoV94 · 2024-06-27T09:15:38Z

@Armavica may be crazy work, but can we get a separate commit where we make the non-strict zips. That way it's easier to evaluate if it sounds correct or may be a bug somewhere?

Or I guess I can just ctrl+f for it

Armavica · 2024-06-28T07:27:11Z

@ricardoV94 Yes, I was planning to present this PR(s) in several steps:

Commits that add strict=True to zips without tests failing
Commits that add strict=True to zips and fix their failures (bugs)
Commits that add strict=False to the remaining zips that need it, or rewrite them so they can be made strict
How does that sound to you?

ricardoV94 · 2024-06-28T09:54:42Z

Sounds good @Armavica

codecov · 2024-06-29T06:56:22Z

Codecov Report

Attention: Patch coverage is 91.70306% with 19 lines in your changes missing coverage. Please review.

Project coverage is 81.05%. Comparing base (05d376f) to head (e3965ef).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #850      +/-   ##
==========================================
+ Coverage   81.04%   81.05%   +0.01%     
==========================================
  Files         170      170              
  Lines       46962    46983      +21     
  Branches    11507    11510       +3     
==========================================
+ Hits        38059    38082      +23     
- Misses       6694     6695       +1     
+ Partials     2209     2206       -3

Files	Coverage Δ
pytensor/compile/builders.py	`88.42% <100.00%> (ø)`
pytensor/compile/function/pfunc.py	`82.92% <100.00%> (ø)`
pytensor/compile/function/types.py	`79.94% <100.00%> (ø)`
pytensor/gradient.py	`77.37% <100.00%> (ø)`
pytensor/graph/basic.py	`88.60% <100.00%> (ø)`
pytensor/graph/op.py	`87.89% <ø> (ø)`
pytensor/graph/replace.py	`84.21% <100.00%> (ø)`
pytensor/graph/rewriting/basic.py	`70.34% <100.00%> (ø)`
pytensor/link/c/basic.py	`87.48% <100.00%> (ø)`
pytensor/link/c/cmodule.py	`56.88% <100.00%> (ø)`
... and 55 more

... and 2 files with indirect coverage changes

jessegrabowski · 2024-07-07T02:29:22Z

Is this ready for review? Seems like all test are passing now

Armavica · 2024-07-07T06:14:17Z

Is this ready for review? Seems like all test are passing now

There are still 11 10 instances of non-strict zips that produce errors if I make them strict, I need to investigate them one by one to see if that's expected behaviour or not.

Armavica · 2024-07-07T13:36:06Z

Actually, I find it difficult to make more progress here, so I am signalling this for review.
I added 464 easy strict=True, 18 less immediate ones, and there are still 10 strict=False that I find the most difficult to understand. I think that they could be handled in another PR. This one introduces 464+18 = 482 safeguards, which I think is a good score :)

ricardoV94 · 2024-07-08T13:56:33Z

pytensor/scalar/loop.py

@@ -93,7 +93,7 @@ def _validate_updates(
                )
        else:
            update = outputs
-        for i, u in zip(init, update, strict=False):
+        for i, u in zip(init[: len(update)], update, strict=True):


Can we use strict=False here?

ricardoV94 · 2024-07-08T13:57:38Z

tests/tensor/conv/test_abstract_conv.py

@@ -1745,7 +1745,7 @@ def setup_method(self):
        self.random_stream = np.random.default_rng(utt.fetch_seed())

        self.inputs_shapes = [(8, 1, 12, 12), (1, 1, 5, 5), (1, 1, 5, 6), (1, 1, 6, 6)]
-        self.filters_shapes = [(5, 1, 2, 2), (1, 1, 3, 3)]
+        self.filters_shapes = [(5, 1, 2, 2), (1, 1, 3, 3)] * 2


Was this a bug?

Yes, it was zipping input_shapes and filter_shapes together so the two last input_shapes were never being used in the tests.

ricardoV94 · 2024-07-08T13:59:33Z

pytensor/tensor/rewriting/subtensor.py

@@ -648,7 +648,7 @@ def local_subtensor_of_alloc(fgraph, node):
    # Slices to take from val
    val_slices = []

-    for i, (sl, dim) in enumerate(zip(slices, dims, strict=False)):
+    for i, (sl, dim) in enumerate(zip(slices, dims[: len(slices)], strict=True)):


Like my previous comment, I find this less readable. The strict=False indicates clearly that we don't expect the sequences to necessarily have the same length?

But if they're not the same length why are we zipping them? Are we sure they're always ordered correctly?

Because they are ordered correctly yes, and presumably what comes after doesn't matter. It's quite common in Subtensor operations / rewrites

I don't mind reverting this, but just to argue a bit in the favor of strict=True, I think an additional advantage of this approach is that it makes it clearer which one of the two lists is supposed to be shorter. I personally find that I understand more about what is happening here when I read this version compared to strict=False.

ricardoV94 · 2024-07-08T14:02:33Z

pytensor/tensor/shape.py

+    if len(shape) != x.type.ndim:
+        return _specify_shape(x, *shape)
+
+    new_shape_matches = all(
+        s == xts for (s, xts) in zip(shape, x.type.shape, strict=True) if s is not None
+    )
+    if new_shape_matches:


This is awkward, the use of strict=False seems fine

I am surprised, this really looks better to me. What do you think of:

if len(shape) != x.type.ndim: return _specify_shape(x, *shape) if all(s in (None, xts) for (s, xts) in zip(shape, x.type.shape, strict=True)):

My problem is the double call to SpecifyShape.

Also, we already established in the comment that if there's different lengths the function is going to raise so the strict=False follows naturally?

Armavica force-pushed the zip-strict branch 2 times, most recently from ada9880 to a5de1b6 Compare June 26, 2024 22:40

Armavica force-pushed the zip-strict branch 5 times, most recently from dc0aa6e to 36868a5 Compare June 29, 2024 06:32

Armavica force-pushed the zip-strict branch from ee484de to 1218135 Compare July 3, 2024 09:31

Armavica force-pushed the zip-strict branch 2 times, most recently from 5746335 to e3965ef Compare July 7, 2024 13:05

Armavica marked this pull request as ready for review July 7, 2024 13:36

Armavica added maintenance no releasenotes labels Jul 7, 2024

Armavica requested a review from jessegrabowski July 8, 2024 10:32

ricardoV94 reviewed Jul 8, 2024

View reviewed changes

ricardoV94 mentioned this pull request Jul 8, 2024

Make zip strict #840

Closed

Armavica force-pushed the zip-strict branch 2 times, most recently from 6a15a26 to 9007ce1 Compare July 9, 2024 14:50

Armavica force-pushed the zip-strict branch from 9007ce1 to c58bf33 Compare July 23, 2024 17:46

Add a strict argument to all zips

765b30f

Armavica added 11 commits November 17, 2024 00:00

Enable the ruff rule ensuring explicit strictness for zips

b22f5e0

Make non-strict zips strict in tests/scan

dc63332

Make non-strict zips strict in tensor/elemwise_cgen

5b95cf5

Make non-strict zip strict in scalar/loop.py

25cfa02

Make non-strict zip strict in printing.py

ab5d6ea

Make non-strict zip strict in test_abstract_conv

eca57c3

Rewrite local_merge_alloc to remove a non-strict zip

32586d1

Make non-strict zip strict in tensor/random/utils

d1a6ad1

Make non-strict zip strict in local_subtensor_of_alloc

267e32c

Make non-strict zip strict in tensor/subtensor.py

fe7018c

Make non-strict zip strict in tensor/shape.py

6124822

Armavica force-pushed the zip-strict branch from c58bf33 to 6124822 Compare November 16, 2024 23:01

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Make all zips explicitly strict or non-strict #850

Make all zips explicitly strict or non-strict #850

Armavica commented Jun 24, 2024 •

edited

Loading

ricardoV94 commented Jun 27, 2024 •

edited

Loading

Armavica commented Jun 28, 2024 •

edited

Loading

ricardoV94 commented Jun 28, 2024

codecov bot commented Jun 29, 2024 •

edited

Loading

jessegrabowski commented Jul 7, 2024

Armavica commented Jul 7, 2024 •

edited

Loading

Armavica commented Jul 7, 2024

ricardoV94 Jul 8, 2024

ricardoV94 Jul 8, 2024

Armavica Jul 8, 2024

ricardoV94 Jul 8, 2024

jessegrabowski Jul 8, 2024

ricardoV94 Jul 8, 2024

Armavica Jul 8, 2024 •

edited

Loading

ricardoV94 Jul 8, 2024 •

edited

Loading

Armavica Jul 9, 2024

ricardoV94 Jul 9, 2024

Make all zips explicitly strict or non-strict #850

Are you sure you want to change the base?

Make all zips explicitly strict or non-strict #850

Conversation

Armavica commented Jun 24, 2024 • edited Loading

Description

Related Issue

Checklist

Type of change

ricardoV94 commented Jun 27, 2024 • edited Loading

Armavica commented Jun 28, 2024 • edited Loading

ricardoV94 commented Jun 28, 2024

codecov bot commented Jun 29, 2024 • edited Loading

Codecov Report

jessegrabowski commented Jul 7, 2024

Armavica commented Jul 7, 2024 • edited Loading

Armavica commented Jul 7, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Armavica Jul 8, 2024 • edited Loading

Choose a reason for hiding this comment

ricardoV94 Jul 8, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Armavica commented Jun 24, 2024 •

edited

Loading

ricardoV94 commented Jun 27, 2024 •

edited

Loading

Armavica commented Jun 28, 2024 •

edited

Loading

codecov bot commented Jun 29, 2024 •

edited

Loading

Armavica commented Jul 7, 2024 •

edited

Loading

Armavica Jul 8, 2024 •

edited

Loading

ricardoV94 Jul 8, 2024 •

edited

Loading