Add support for qml.specs to the beta QNode #1739

josh146 · 2021-10-12T09:49:18Z

Context: Previous, qml.specs would query QNode.specs. This property does not exist on the new QNode.

Description of the Change:

qml.specs transform is modified to directly compute the specifications of qml.beta.QNode objects.
An if-else block used to temporarily set the QNode expansion depth is changed to try-finally, to ensure it is safe in case there is an exception.

Benefits:

The beta QNode correctly tracks trainable parameters in backpropagation mode
Computing the number of gradient executions is now more reliable, since the gradient transform is simply called directly.

Possible Drawbacks: n/a

Related GitHub Issues: n/a

github-actions · 2021-10-12T09:49:36Z

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.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.

josh146 · 2021-10-12T09:55:39Z

[sc-9747]

codecov · 2021-10-12T10:05:24Z

Codecov Report

Merging #1739 (655e1c1) into master (65bdcfc) will increase coverage by 2.66%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1739      +/-   ##
==========================================
+ Coverage   96.55%   99.22%   +2.66%     
==========================================
  Files         207      207              
  Lines       15552    15573      +21     
==========================================
+ Hits        15017    15452     +435     
+ Misses        535      121     -414

Impacted Files	Coverage Δ
pennylane/gradients/parameter_shift.py	`100.00% <100.00%> (ø)`
pennylane/transforms/specs.py	`100.00% <100.00%> (ø)`
pennylane/interfaces/batch/__init__.py	`100.00% <0.00%> (+0.96%)`	⬆️
pennylane/devices/default_qubit.py	`100.00% <0.00%> (+1.22%)`	⬆️
pennylane/beta/devices/default_tensor.py	`96.93% <0.00%> (+1.70%)`	⬆️
pennylane/interfaces/batch/tensorflow.py	`100.00% <0.00%> (+2.22%)`	⬆️
pennylane/devices/default_qubit_tf.py	`92.00% <0.00%> (+2.66%)`	⬆️
pennylane/beta/devices/default_tensor_tf.py	`90.62% <0.00%> (+3.12%)`	⬆️
pennylane/interfaces/batch/torch.py	`100.00% <0.00%> (+3.27%)`	⬆️
pennylane/interfaces/torch.py	`100.00% <0.00%> (+3.29%)`	⬆️
... and 16 more

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 65bdcfc...655e1c1. Read the comment docs.

pennylane/transforms/specs.py

…nto beta-qnode-specs

josh146 · 2021-10-12T16:04:54Z

@albi3ro I've updated the test file to retain the original tests and the new tests, ensuring qml.specs continues to work well with both QNodes 🙂

albi3ro

Just some thoughts and questions for now.

albi3ro · 2021-10-12T13:23:55Z

pennylane/transforms/specs.py

+        info["execution_options"] = qnode.execute_kwargs
+        info["interface"] = qnode.interface
+
+        if callable(qnode.gradient_fn):


What else would it be? A string? None?

Yep, it can currently take three types:

Callable: one of qml.gradients.gradient_transform (could be changed to isinstance(qnode.gradient_fn, gradient_transform) to be stricter)

None: no gradients

str: to indicate "device"

pennylane/transforms/specs.py

albi3ro · 2021-10-12T13:31:23Z

pennylane/transforms/specs.py

+                "The QNode specifications can only be calculated after its quantum tape has been constructed."
+            )
+
+        info = qnode.qtape.specs.copy()


Will we be getting rid of the qnode.specs property?

Yep; the idea is to remove all properties/methods that depend on the state of construction, and replace them with functions that act on the QNode. This also includes qnode.draw() and qnode.metric_tensor().

albi3ro · 2021-10-12T13:40:04Z

pennylane/transforms/specs.py

+                # In the case of a broad exception, we don't want the `qml.specs` transform
+                # to fail. Instead, we simply indicate that the number of gradient executions
+                # is not supported.


While this might be helpful while the new QNode is still under development, in the long run, we should know exactly the types of situations that would cause this to fail and check for those instead. If something's going wrong internally, we want to know.

Actually, this was inserted to solve an error I was getting while running the tests! In particular, this one:

@qml.qnode(dev, diff_method=diff_method) def circuit(): return qml.state()

Of the three diff methods tested, only backprop supports state differentiation; both adjoint and parameter-shift will raise an error.

Previously, qml.specs would use the Operator.grad_recipe to determine the number of shifts required, independent of the circuit measurement, resulting in an incorrect value for num_gradient_executions.

In this PR, qml.specs queries directly the gradient transform to determine the number of shifts. This results in more accurate results, however it means that calling qml.specs(circuit) on the QNode above will raise an error, since qml.gradients.param_shift and adjoint raise an exception if the input circuit ends with qml.state.

Overall, I think this is more informative; checking the number of parameter-shift executions on this QNode will show NotSupported, rather than a numeric value.

What we could do, though, is somehow 'catch' the exception message, and store this in the dictionary?

I've pushed an update that does the following 🙂

dev = qml.device("default.qubit", wires=2) @qml.beta.qnode(dev, diff_method=qml.gradients.param_shift) def circuit(x): qml.RX(x, wires=0) return qml.state()

>>> specs = qml.specs(circuit)(0.56) >>> specs["num_gradient_executions"] NotSupported: Computing the gradient of circuits that return the state is not supported.

pennylane/gradients/parameter_shift.py

albi3ro · 2021-10-14T12:57:26Z

pennylane/transforms/specs.py

+            info["gradient_fn"] = inspect.getmodule(qnode.gradient_fn).__name__
+
+            try:
+                if isinstance(qnode.gradient_fn, qml.gradients.gradient_transform):


This seems to already be inside an if callable(qnode.gradient_fn) block. Is there a reason we make the stricter check again? Or can we defer this check to the outer if statement?

Good point, let me remove the inner one.

albi3ro

A few questions about when:
A) someone passes in a gradient transform that's built into pennylane, like qml.gradients.param_shift
B) someone passes in their own custom gradient transform

I think those cases need to be improved and tested.

pennylane/transforms/specs.py

Co-authored-by: Christina Lee <christina@xanadu.ai>

josh146 · 2021-10-14T15:18:22Z

Thanks Christina, both valid points! I've pushed a commit that takes this into account, and now shows the absolute import path for gradient transforms, even when a custom transform.

albi3ro

Thanks for the improvements!

Add support for qml.specs to the beta QNode

bda09ea

josh146 added 3 commits October 12, 2021 17:52

changelog

d873106

changelog

11959ca

changelog

e278e4e

josh146 requested a review from albi3ro October 12, 2021 09:55

josh146 added the review-ready 👌 PRs which are ready for review by someone from the core team. label Oct 12, 2021

albi3ro reviewed Oct 12, 2021

View reviewed changes

pennylane/transforms/specs.py Outdated Show resolved Hide resolved

josh146 added 4 commits October 13, 2021 00:00

keep old tests

b28c918

Merge branch 'master' into beta-qnode-specs

9b91506

suggested test

d0c7989

Merge branch 'beta-qnode-specs' of github.com:PennyLaneAI/pennylane i…

6ea474a

…nto beta-qnode-specs

albi3ro reviewed Oct 12, 2021

View reviewed changes

josh146 added 3 commits October 13, 2021 14:41

Merge branch 'master' into beta-qnode-specs

2f6ac26

suggested changes

5a617d6

suggested changes

120e3b7

josh146 requested a review from albi3ro October 13, 2021 07:02

add another test

78a2a37

josh146 commented Oct 13, 2021

View reviewed changes

pennylane/gradients/parameter_shift.py Show resolved Hide resolved

josh146 added 2 commits October 14, 2021 00:41

Update pennylane/gradients/parameter_shift.py

2640ec4

Update parameter_shift.py

731d6fa

albi3ro reviewed Oct 14, 2021

View reviewed changes

albi3ro requested changes Oct 14, 2021

View reviewed changes

pennylane/transforms/specs.py Outdated Show resolved Hide resolved

pennylane/transforms/specs.py Outdated Show resolved Hide resolved

josh146 and others added 4 commits October 14, 2021 22:56

Update specs.py

94dc46f

Update pennylane/transforms/specs.py

701e790

Co-authored-by: Christina Lee <christina@xanadu.ai>

Merge branch 'master' into beta-qnode-specs

c7cf9c8

suggested changes

655e1c1

josh146 requested a review from albi3ro October 14, 2021 15:18

albi3ro approved these changes Oct 14, 2021

View reviewed changes

josh146 merged commit 7425c5d into master Oct 14, 2021

josh146 deleted the beta-qnode-specs branch October 14, 2021 16:03

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for qml.specs to the beta QNode #1739

Add support for qml.specs to the beta QNode #1739

josh146 commented Oct 12, 2021

github-actions bot commented Oct 12, 2021

josh146 commented Oct 12, 2021

codecov bot commented Oct 12, 2021 •

edited

Loading

josh146 commented Oct 12, 2021

albi3ro left a comment

albi3ro Oct 12, 2021

josh146 Oct 13, 2021

albi3ro Oct 12, 2021

josh146 Oct 13, 2021

albi3ro Oct 12, 2021

josh146 Oct 13, 2021

josh146 Oct 13, 2021

josh146 Oct 13, 2021 •

edited

Loading

albi3ro Oct 14, 2021

josh146 Oct 14, 2021

albi3ro left a comment

josh146 commented Oct 14, 2021

albi3ro left a comment

Add support for qml.specs to the beta QNode #1739

Add support for qml.specs to the beta QNode #1739

Conversation

josh146 commented Oct 12, 2021

github-actions bot commented Oct 12, 2021

josh146 commented Oct 12, 2021

codecov bot commented Oct 12, 2021 • edited Loading

Codecov Report

josh146 commented Oct 12, 2021

albi3ro left a 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

josh146 Oct 13, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

albi3ro left a comment

Choose a reason for hiding this comment

josh146 commented Oct 14, 2021

albi3ro left a comment

Choose a reason for hiding this comment

codecov bot commented Oct 12, 2021 •

edited

Loading

josh146 Oct 13, 2021 •

edited

Loading