Add function to apply instantiated operations to queuing contexts #1433

josh146 · 2021-06-24T09:20:24Z

Context:

Currently, there is no consistent, user-facing approach for applying (already instantiated) operators to a queuing context. Instead, there is a combination of different approaches:

op.queue(): the most common approach. Almost all operators provide their own queuing logic currently, bar Tensor, which inherits a basic implementation from the base class and leads to inconsistent behaviour.
qml.QueuingContext.append(): appends the object directly to the currently active queuing context.
context._append() appends the object to a specific queuing context.

This situation is not ideal, since op.queue() is not consistent, not documented, and not user-facing.

Description of the Change:

Adds Tensor.queue() to make the Tensor class consistent with other Operator subclasses. This results in several accidental bug fixes that have plagued the Hamiltonian class (see Adds Hamiltonian Queueing and Transform to Compute Hamiltonian Expectation Value #1142)
Modifies the existing queue() methods, so that they now take an optional context keyword argument. This allows operators to be queued to specific, provided, contexts, rather than just the currently active context.
Adds a new function qml.apply, which will be the new user and developer facing method for queuing already instantiated operations.

Benefits:

There is now a consistent, documented, and well-tested method for queuing instantiated operators to queuing contexts.

Possible Drawbacks:

The existing qml.apply function was an import of qml.collections.apply. This function was infrequently used or advertised, and now must be accessed by importing directly from the collections module.

Related GitHub Issues: n/a

github-actions · 2021-06-24T09:20:41Z

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.

codecov · 2021-06-24T10:20:02Z

Codecov Report

Merging #1433 (0593611) into master (580172a) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1433   +/-   ##
=======================================
  Coverage   98.24%   98.24%           
=======================================
  Files         160      160           
  Lines       11991    12016   +25     
=======================================
+ Hits        11780    11805   +25     
  Misses        211      211

Impacted Files	Coverage Δ
pennylane/__init__.py	`98.57% <100.00%> (ø)`
pennylane/measure.py	`91.86% <100.00%> (+0.09%)`	⬆️
pennylane/operation.py	`95.92% <100.00%> (+0.07%)`	⬆️
pennylane/queuing.py	`97.64% <100.00%> (+0.54%)`	⬆️
pennylane/vqe/vqe.py	`92.19% <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 580172a...0593611. Read the comment docs.

tests/test_queuing.py

josh146 · 2021-06-24T12:31:43Z

tests/test_vqe.py

+            qml.PauliZ(0),
+            qml.PauliZ(2),


There are two tests in this test function; one that tests Hamiltonian queuing for a Hamiltonian created outside the context, and one that tests Hamiltonian queuing for a Hamiltonian created inside the context.

This variable queue is the expected result for the Hamiltonian outside the context. If you scroll down, you can see that the addition of these two elements makes it consistent with the expected result for the Hamiltonian inside the context.

.github/CHANGELOG.md

glassnotes

Just took a quick look at the code for now, will provide more comments after using this for some local tests 👍

.github/CHANGELOG.md

pennylane/queuing.py

tests/test_queuing.py

josh146 · 2021-06-29T12:47:21Z

@glassnotes @mariaschuld I have renamed the function to qml.apply() 🙂

pennylane/operation.py

glassnotes · 2021-06-29T17:25:04Z

pennylane/queuing.py

+                qml.CNOT(wires=[0, 1])
+
+        >>> tape1.operations
+        [Hadamard(wires=[1]), <QuantumTape: wires=[0], params=1>, PauliX(wires=[0]), CNOT(wires=[0, 1])]


Why is the <QuantumTape: wires=[0], params=1> part there? It looks like only a Hadamard, PauliX, and CNOT should be applied to tape1?

Oh wait I think I just understood why from the tests later on; the contents of tape2 are also getting added to tape1 because it's within the same queuing context, but using qml.apply allows us to apply something to tape1 without also applying it to tape2?

Yep exactly!

glassnotes · 2021-06-29T17:36:04Z

pennylane/queuing.py

+                qml.Hadamard(wires=1)
+
+                with qml.tape.QuantumTape() as tape2:
+                    op1 = qml.PauliX(wires=0)


This behaviour actually doesn't feel so intuitive; if I create something and assign it to a variable with the intention of queuing it in a different context, I wouldn't expect it to also be queued in the context that created it. Maybe the fact that operations are queued whenever they are instantiated could just be bolded or otherwise emphasized above at the start of the Example section, and be explicit that this includes cases like this one?

I will emphasize it 👍 I think the example here is very much 'advanced' --- day-to-day, I think by the far the most common use-case of this functionality is to queue an operation that was instantiated outside of any queuing context.

Great suggestion @glassnotes; I've attempted to make it more clear here: 13aaea2

That looks awesome! 💯

pennylane/queuing.py

Co-authored-by: Olivia Di Matteo <2068515+glassnotes@users.noreply.github.com>

mariaschuld · 2021-06-30T12:23:54Z

pennylane/operation.py

-                self.obs.append(o)
-            else:
-                raise ValueError("Can only perform tensor products between observables.")
+    def queue(self, context=qml.QueuingContext, init=False):  # pylint: disable=arguments-differ


I'm not sure I understand the init argument?

Ah - the queuing logic was previous occuring at the exact same time as the Tensor datastructure was being computed.

That is, within __init__, there was a single for loop that did two things:

It queued each constituent observable

It created the internal obs list of constituent observables.

I originally split it into two separate for loops in order to separate out the logic, but realized that it would result in looping through the observable list twice. Which is fine... but as the number of qubits increases, this could be a bottleneck.

So I combined it back into a single for loop, but use init=True to ensure that building the datastructure only happens on instantiation, while queuing happens every time.

mariaschuld · 2021-06-30T12:24:27Z

pennylane/queuing.py

@@ -220,6 +221,9 @@ def _append(self, obj, **kwargs):
    def _remove(self, obj):
        self.queue.remove(obj)

+    append = _append
+    remove = _remove


Trying to understand this one?

Ah, this one is weird. QueuingContext.append is a class method, so doesn't take self into account. Here, we are in a subclass, and simply overriding the inherited class methods.

pennylane/queuing.py

mariaschuld · 2021-06-30T12:28:30Z

tests/test_queuing.py

+        """Test that applying an operation without an active
+        context raises an error"""
+        with pytest.raises(RuntimeError, match="No queuing context"):
+            qml.apply(qml.PauliZ(0))


mariaschuld · 2021-06-30T12:28:59Z

tests/test_queuing.py

+            op1 = qml.PauliZ(0)
+            op2 = qml.apply(op1)
+
+        assert tape.operations == [op1, op2]


also cool, makes total sense!

mariaschuld

Very nice Josh! I have to admit I don't follow all changes at the beginning, but the main function, docstring and tests look awesome!

glassnotes

@josh146 thanks for updating the documentation, helps make the examples much clearer! I've incorporated the method into my WIP transforms, and everything works seamlessly 💯

pennylane/queuing.py

Add function to apply instantiated operations to queuing contexts

fea2b6f

josh146 added the WIP 🚧 Work-in-progress label Jun 24, 2021

josh146 added 7 commits June 24, 2021 17:21

fixes

6b8e2fd

linting

8b7abbe

fix tests

cd7fb7e

fix tests

cf5fb68

linting

4a792f0

fixes

bba80d8

fixes

07e9edb

josh146 commented Jun 24, 2021

View reviewed changes

tests/test_queuing.py Show resolved Hide resolved

add tests

535c82e

josh146 commented Jun 24, 2021

View reviewed changes

josh146 added 2 commits June 24, 2021 20:45

add tests

a784aa4

add documentation

693a61b

josh146 requested a review from co9olguy June 24, 2021 13:26

josh146 removed the WIP 🚧 Work-in-progress label Jun 24, 2021

josh146 changed the title ~~[WIP] Add function to apply instantiated operations to queuing contexts~~ Add function to apply instantiated operations to queuing contexts Jun 24, 2021

josh146 requested a review from glassnotes June 24, 2021 13:26

josh146 commented Jun 24, 2021

View reviewed changes

.github/CHANGELOG.md Show resolved Hide resolved

josh146 added the review-ready 👌 PRs which are ready for review by someone from the core team. label Jun 24, 2021

glassnotes reviewed Jun 24, 2021

View reviewed changes

.github/CHANGELOG.md Show resolved Hide resolved

pennylane/queuing.py Outdated Show resolved Hide resolved

pennylane/queuing.py Show resolved Hide resolved

tests/test_queuing.py Show resolved Hide resolved

Rename to qml.apply

ac6e336

josh146 requested review from glassnotes and mariaschuld June 29, 2021 12:46

Merge branch 'master' into apply

ec77ce6

mariaschuld reviewed Jun 29, 2021

View reviewed changes

pennylane/operation.py Show resolved Hide resolved

glassnotes reviewed Jun 29, 2021

View reviewed changes

josh146 and others added 3 commits June 30, 2021 11:31

Update pennylane/queuing.py

a48c8cd

Co-authored-by: Olivia Di Matteo <2068515+glassnotes@users.noreply.github.com>

Merge branch 'master' into apply

53b8071

suggested changes

13aaea2

mariaschuld reviewed Jun 30, 2021

View reviewed changes

pennylane/queuing.py Outdated Show resolved Hide resolved

mariaschuld reviewed Jun 30, 2021

View reviewed changes

mariaschuld approved these changes Jun 30, 2021

View reviewed changes

glassnotes approved these changes Jun 30, 2021

View reviewed changes

josh146 commented Jun 30, 2021

View reviewed changes

pennylane/queuing.py Show resolved Hide resolved

josh146 commented Jun 30, 2021

View reviewed changes

pennylane/queuing.py Show resolved Hide resolved

josh146 added 2 commits June 30, 2021 22:24

Apply suggestions from code review

80940d8

Merge branch 'master' into apply

0593611

josh146 merged commit fe71ae0 into master Jun 30, 2021

josh146 deleted the apply branch June 30, 2021 16:14

glassnotes mentioned this pull request Jul 7, 2021

Unable to directly queue operations multiple times for tape transform #1218

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add function to apply instantiated operations to queuing contexts #1433

Add function to apply instantiated operations to queuing contexts #1433

josh146 commented Jun 24, 2021 •

edited

Loading

github-actions bot commented Jun 24, 2021

codecov bot commented Jun 24, 2021 •

edited

Loading

josh146 Jun 24, 2021

glassnotes left a comment

josh146 commented Jun 29, 2021

glassnotes Jun 29, 2021

glassnotes Jun 29, 2021

josh146 Jun 30, 2021

glassnotes Jun 29, 2021

josh146 Jun 30, 2021

josh146 Jun 30, 2021

glassnotes Jun 30, 2021

mariaschuld Jun 30, 2021

josh146 Jun 30, 2021

mariaschuld Jun 30, 2021

josh146 Jun 30, 2021

mariaschuld Jun 30, 2021

mariaschuld Jun 30, 2021

mariaschuld left a comment

glassnotes left a comment

Add function to apply instantiated operations to queuing contexts #1433

Add function to apply instantiated operations to queuing contexts #1433

Conversation

josh146 commented Jun 24, 2021 • edited Loading

github-actions bot commented Jun 24, 2021

codecov bot commented Jun 24, 2021 • edited Loading

Codecov Report

Choose a reason for hiding this comment

glassnotes left a comment

Choose a reason for hiding this comment

josh146 commented Jun 29, 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

Choose a reason for hiding this comment

mariaschuld left a comment

Choose a reason for hiding this comment

glassnotes left a comment

Choose a reason for hiding this comment

josh146 commented Jun 24, 2021 •

edited

Loading

codecov bot commented Jun 24, 2021 •

edited

Loading