[Templates] Rewrite layer templates as operations #1163

mariaschuld · 2021-03-25T10:19:11Z

Porting layer templates to operations.

Things that were changed are:

docstrings improved
broadcast unpacked in every decomposition
shape method added (and mentioned in docstring)
tests are put in separate file and largely rewritten

There are 2 outstanding issues:

Since CVNeuralNetLayers depends on Interferometer I thought I also port this template quickly, but there was a problem with the existing gate! I uncommented the really bad test case for now, and another one is failing.
@Thenerdstation is fixing a problem with jax differentiation in ParticleConservingU1

Otherwise ready to review. Sorry, I know it is a nightmare because so much code moved, but just a sanity check regarding the tests and the structure (as well as docstrings) would be great!

github-actions · 2021-03-25T10:19:30Z

Hello. You may have forgotten to update the changelog!
Please edit .github/CHANGELOG.md with:

A one-to-two sentence description of the change. You may include a small working example for new features.
A link back to this PR.
Your name (or GitHub username) in the contributors section.

mariaschuld · 2021-03-25T15:49:15Z

tests/templates/test_layers/test_cv_neural_net.py

+    return qml.expval(qml.X(0))
+
+
+# def circuit_decomposed(*weights):


This has to be uncommented once the interferometer issue is solved

Reminder to uncomment!

Ah, thanks, I forgot!

mariaschuld · 2021-03-25T15:49:58Z

tests/templates/test_layers/test_particle_conserving_u2.py

+            ),
+            (np.array([1, 1]), np.array([0.0 + 0.0j, 0.0 + 0.0j, 0.0 + 0.0j, 1.0 + 0.0j])),
+        ],
+    )


Todo: As far as I know we discourage test cases that were just run, and then compared to the output...better to have logical sanity checks?

ixfoduap

Having a hard time reviewing this PR with so many changes. Will get back to it with a clearer mind

pennylane/templates/layers/cv_neural_net.py

ixfoduap · 2021-03-26T15:38:04Z

pennylane/templates/layers/cv_neural_net.py

-        broadcast(unitary=Displacement, pattern="single", wires=wires, parameters=a_and_phi_a)
+    @staticmethod
+    def shape(n_layers, n_wires):
+        r"""Returns a list of shapes for the 11 parameter tensors.


tests/templates/test_layers/test_cv_neural_net.py

…o rewrite-layers

codecov · 2021-04-06T08:36:48Z

Codecov Report

Merging #1163 (ace1680) into master (779f9ec) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1163   +/-   ##
=======================================
  Coverage   98.11%   98.12%           
=======================================
  Files         145      145           
  Lines       10943    10971   +28     
=======================================
+ Hits        10737    10765   +28     
  Misses        206      206

Impacted Files	Coverage Δ
pennylane/templates/layers/basic_entangler.py	`100.00% <ø> (ø)`
pennylane/templates/layers/cv_neural_net.py	`100.00% <100.00%> (ø)`
...nnylane/templates/layers/particle_conserving_u1.py	`100.00% <100.00%> (ø)`
...nnylane/templates/layers/particle_conserving_u2.py	`100.00% <100.00%> (ø)`
pennylane/templates/layers/random.py	`100.00% <100.00%> (ø)`
...ennylane/templates/layers/simplified_two_design.py	`100.00% <100.00%> (ø)`
pennylane/templates/layers/strongly_entangling.py	`100.00% <100.00%> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 779f9ec...ace1680. Read the comment docs.

mariaschuld · 2021-04-06T09:57:54Z

@ixfoduap and @Thenerdstation I know this is a boring PR, but would you mind having a general look? :)

pennylane/templates/layers/basic_entangler.py

pennylane/templates/layers/particle_conserving_u1.py

chaserileyroberts · 2021-04-06T15:10:14Z

pennylane/templates/layers/random.py

+    def expand(self):
+
+        if self.seed is not None:
+            np.random.seed(self.seed)
+
+        shape = qml.math.shape(self.parameters[0])
+
+        with qml.tape.QuantumTape() as tape:
+
+            for l in range(self.n_layers):
+
+                i = 0
+                while i < shape[1]:
+                    if np.random.random() > self.ratio_imprimitive:
+                        # apply a random rotation gate to a random wire
+                        gate = np.random.choice(self.rotations)
+                        rnd_wire = self.wires.select_random(1)
+                        gate(self.parameters[0][l, i], wires=rnd_wire)
+                        i += 1
+
+                    else:
+                        # apply the entangler to two random wires
+                        if len(self.wires) > 1:
+                            rnd_wires = self.wires.select_random(2)
+                            self.imprimitive(wires=rnd_wires)
+        return tape


Do we care at all that expanding the tape will give different circuits each time?

That's the idea of this template...to avoid, the user must feed a seed.

chaserileyroberts · 2021-04-06T15:19:14Z

pennylane/templates/layers/strongly_entangling.py

-        strongly_entangling_layer(
-            weights=weights[l], wires=wires, r=ranges[l], imprimitive=imprimitive
-        )
+        return n_layers, n_wires, 3


Suggested change

return n_layers, n_wires, 3

return (n_layers, n_wires, 3)

Doesn't matter at all but makes it more clear to me it should be treated as a tuple and not 3 separate return args.

Codefactor complains about the unnecessary brakets though...?

chaserileyroberts · 2021-04-06T15:19:38Z

tests/templates/test_integration.py

-QUBIT_DIFFABLE_NONDIFFABLE = [(qml.templates.AmplitudeEmbedding,
-                               {'features': [1 / 2, 1 / 2, 1 / 2, 1 / 2]},
-                               {'wires': [0, 1], 'normalize': False},
-                               2),
-                              (qml.templates.AmplitudeEmbedding,
-                               {'features': [1 / 2, 1 / 2, 1 / 2, 1 / 2]},
-                               {'wires': [0, 1], 'normalize': True},
-                               2),
-                              (qml.templates.BasisEmbedding,
-                               {},
-                               {'wires': [0, 1], 'features': [1, 0]},
-                               2),


Any reason you removed most of these?

I was a bit inconsistent here...the test_integration file will be deleted once all refactor PRs are merged, because the old tests are superseeded. I first deleted things from this file in each PR, then realised I can just leave it and delete in the end. So I'll take care of this!

chaserileyroberts · 2021-04-06T15:22:05Z

tests/templates/test_layers/test_basic_entangler.py


-        assert type(tape.operations[0]) == rotation
-        assert type(tape.operations[1]) == rotation
+        assert rotation in [type(gate) for gate in queue]


This doesn't check that all gates are rotations, just that a rotation exists in that list.

Can you instead do assert all(type(gate) == rotation for gate in queue)?

But the queue does not only have rotations...I basically just want to check that the custom gate type is used here...where it is used is tested in the first test...

tests/templates/test_layers/test_basic_entangler.py

chaserileyroberts · 2021-04-06T15:24:54Z

tests/templates/test_layers/test_basic_entangler.py

+
+        res = circuit(weights)
+        res2 = circuit2(weights)
+        assert qml.math.allclose(res, res2, atol=tol, rtol=0)


Suggested change

assert qml.math.allclose(res, res2, atol=tol, rtol=0)

np.testing.assert_allclose(res, res2, atol=tol, rtol=0)

This works with TF/Torch/JAX tensor-types too!

I just checked and works here, but not when a tensor has a grad attached...so will not change it everywhere...

chaserileyroberts · 2021-04-06T15:31:28Z

tests/templates/test_layers/test_particle_conserving_u2.py

+        circuit()
+        circuit2()
+
+        assert np.allclose(dev.state, dev2.state, atol=tol, rtol=0)


Can we instead have the circuits return qml.state() and check that output instead of relying on dev.state? In the next quarter, we're going to start enforcing devices be stateless objects/functions, so this will be one of the main things that will be deprecated.

Sorry, I should have left a comment about this (did so in the other PRs): Custom wire labels do not work with qml.state(). So maybe when this gets deprecated, this test is a good motivation to also fix this limitation. So I'd suggest to leave it like this for now?

tests/templates/test_layers/test_simplified_twodesign.py

ixfoduap

I have to admit that my head hurts trying to review this PR. I'm requesting changes to ensure that Chase's comments are addressed and that a section of the code is uncommented with the resolved interferometer issue.

Otherwise I'm willing to approve to not slow down the process, but I do so without confidence in the correctness of the changes

pennylane/templates/layers/basic_entangler.py

ixfoduap · 2021-04-07T21:54:27Z

tests/templates/test_layers/test_cv_neural_net.py

+    return qml.expval(qml.X(0))
+
+
+# def circuit_decomposed(*weights):


Reminder to uncomment!

Co-authored-by: Chase Roberts <chase@xanadu.ai>

mariaschuld · 2021-04-08T06:07:14Z

Thanks so much for the reviews, I know they are hard, but you already picked up a few crucial oversights!

Ready for final check!

ixfoduap

YOLO approve

mariaschuld added 3 commits March 24, 2021 12:56

converted CVQNN

8a919b6

backup

6809572

rewrite particle_conserving

f291126

mariaschuld added 6 commits March 25, 2021 15:42

add tests for simplified and stronglyent

0acca6a

add particle1 tests

4d62e4f

finished all tests - 2 external bugs remaining

21827db

update changelog

1431e35

black

892256c

resolve conflict in changelog

558d008

mariaschuld changed the title ~~[Templates] [WIP] Rewrite layer templates as operations~~ [Templates] Rewrite layer templates as operations Mar 25, 2021

mariaschuld requested review from ixfoduap and chaserileyroberts March 25, 2021 15:46

mariaschuld commented Mar 25, 2021

View reviewed changes

delete old tests that are now failing

2b1e639

ixfoduap reviewed Mar 26, 2021

View reviewed changes

mariaschuld added 5 commits March 29, 2021 08:59

Merge branch 'master' into rewrite-layers

73252db

Merge branch 'master' into rewrite-layers

d983b80

implement some review comments

28c3f7e

Merge branch 'rewrite-layers' of github.com:PennyLaneAI/pennylane int…

9f47a4d

…o rewrite-layers

reverse changes to interferometer

d3e748b

improve codecov

292f8fe

mariaschuld mentioned this pull request Apr 6, 2021

[TEMPLATES] Rewrite state preparation templates as operations #1190

Merged

chaserileyroberts reviewed Apr 6, 2021

View reviewed changes

mariaschuld mentioned this pull request Apr 7, 2021

[TEMPLATES] Rewrite subroutines as operations #1192

Merged

ixfoduap requested changes Apr 7, 2021

View reviewed changes

Update tests/templates/test_layers/test_basic_entangler.py

2e5aaf0

Co-authored-by: Chase Roberts <chase@xanadu.ai>

mariaschuld and others added 3 commits April 8, 2021 07:06

Update tests/templates/test_layers/test_basic_entangler.py

7ae4d57

Co-authored-by: Chase Roberts <chase@xanadu.ai>

implement review comments

22358f1

Merge branch 'master' into rewrite-layers

4fcf7c0

mariaschuld requested review from ixfoduap and chaserileyroberts April 8, 2021 06:06

mariaschuld added 4 commits April 8, 2021 08:41

solved merge conflict

481d0b8

more merge conflicts

71b6679

fix changelog

1df440c

delete old integration tests

23caff8

ixfoduap approved these changes Apr 8, 2021

View reviewed changes

mariaschuld added 2 commits April 8, 2021 16:00

Merge branch 'master' into rewrite-layers

7d19cc1

Merge branch 'master' into rewrite-layers

ace1680

mariaschuld merged commit 5645fd2 into master Apr 9, 2021

mariaschuld deleted the rewrite-layers branch April 9, 2021 07:02

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Templates] Rewrite layer templates as operations #1163

[Templates] Rewrite layer templates as operations #1163

mariaschuld commented Mar 25, 2021 •

edited

Loading

github-actions bot commented Mar 25, 2021

mariaschuld Mar 25, 2021

ixfoduap Apr 7, 2021

mariaschuld Apr 8, 2021

mariaschuld Mar 25, 2021

ixfoduap left a comment

ixfoduap Mar 26, 2021

codecov bot commented Apr 6, 2021 •

edited

Loading

mariaschuld commented Apr 6, 2021

chaserileyroberts Apr 6, 2021

mariaschuld Apr 8, 2021

chaserileyroberts Apr 6, 2021

mariaschuld Apr 8, 2021

chaserileyroberts Apr 6, 2021

mariaschuld Apr 8, 2021

chaserileyroberts Apr 6, 2021

mariaschuld Apr 8, 2021

chaserileyroberts Apr 6, 2021

mariaschuld Apr 8, 2021

chaserileyroberts Apr 6, 2021

mariaschuld Apr 8, 2021

ixfoduap left a comment

ixfoduap Apr 7, 2021

mariaschuld commented Apr 8, 2021

ixfoduap left a comment

		return qml.expval(qml.X(0))


		# def circuit_decomposed(*weights):

	assert qml.math.allclose(res, res2, atol=tol, rtol=0)
	np.testing.assert_allclose(res, res2, atol=tol, rtol=0)

[Templates] Rewrite layer templates as operations #1163

[Templates] Rewrite layer templates as operations #1163

Conversation

mariaschuld commented Mar 25, 2021 • edited Loading

github-actions bot commented Mar 25, 2021

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

ixfoduap left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Apr 6, 2021 • edited Loading

Codecov Report

mariaschuld commented Apr 6, 2021

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

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

ixfoduap left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mariaschuld commented Apr 8, 2021

ixfoduap left a comment

Choose a reason for hiding this comment

mariaschuld commented Mar 25, 2021 •

edited

Loading

codecov bot commented Apr 6, 2021 •

edited

Loading