Add transform to insert operations into circuits #1795

trbromley · 2021-10-22T19:19:12Z

Context:

Add insert and insert_in_dev, which can be used to insert operations into fixed points in a circuit (e.g. for the purposes of adding noise):

from pennylane.transforms import insert

dev = qml.device("default.mixed", wires=2)

@qml.qnode(dev)
@insert(qml.AmplitudeDamping, 0.2, position="end")
def f(w, x, y, z):
    qml.RX(w, wires=0)
    qml.RY(x, wires=1)
    qml.CNOT(wires=[0, 1])
    qml.RY(y, wires=0)
    qml.RX(z, wires=1)
    return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1))

f(0.9, 0.4, 0.5, 0.6)  # executions now noisy
print(f.draw())
# 0: ──RX(0.9)──╭C──RY(0.5)──AmplitudeDamping(0.2)──╭┤ ⟨Z ⊗ Z⟩ 
# 1: ──RY(0.4)──╰X──RX(0.6)──AmplitudeDamping(0.2)──╰┤ ⟨Z ⊗ Z⟩

For devices:

from pennylane.beta import qnode

dev = qml.device("default.mixed", wires=4)

@qnode(dev)
def f(w, x, y, z):
    qml.RX(w, wires=0)
    qml.RY(x, wires=1)
    qml.CNOT(wires=[0, 1])
    qml.RY(y, wires=0)
    qml.RX(z, wires=1)
    return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1))

qml.transforms.insert_in_dev(dev, qml.AmplitudeDamping, 0.2)
f(0.9, 0.4, 0.5, 0.6)  # Now noisy

Description of the Change:

Add a new insert_ops submodule to the transforms module containing the above functions, as well as tests and updating docs.

The insert_in_dev prototypes a device transform by overwriting the device's expand_fn.

Benefits:

Users can easily add prototypical noise channels to a quantum circuit.

Possible Drawbacks:

So far the transformation is pretty simple (e.g., operation is added at start, end, or everywhere). If using this to insert noise, it does not closely resemble noise found on hardware.
The device transform is only compatible with the new QNode and new execute UI

Related:

An overrotation type of noise (which preserves purity) was discussed here.

codecov · 2021-10-22T19:48:03Z

Codecov Report

Merging #1795 (96cb3d0) into master (f78bbae) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1795   +/-   ##
=======================================
  Coverage   98.92%   98.92%           
=======================================
  Files         208      209    +1     
  Lines       15655    15708   +53     
=======================================
+ Hits        15486    15539   +53     
  Misses        169      169

Impacted Files	Coverage Δ
pennylane/_device.py	`98.26% <100.00%> (+0.02%)`	⬆️
pennylane/transforms/__init__.py	`100.00% <100.00%> (ø)`
pennylane/transforms/draw.py	`100.00% <100.00%> (ø)`
pennylane/transforms/insert_ops.py	`100.00% <100.00%> (ø)`
pennylane/transforms/qfunc_transforms.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 f78bbae...96cb3d0. Read the comment docs.

trbromley · 2021-10-22T20:06:21Z

pennylane/transforms/noise.py

+    noisy_op_args: Union[tuple, float],
+    position: str = "all",
+) -> QuantumTape:
+    r"""Add noisy operations to an input tape.


Any idea how to fix this? I.e., to have the usual brackets after add_noise_to_tape listing the arguments.

Also this 🤔

Hey Tom, this is all WIP as we understand the best workflow for transforms, but I'm wondering if it makes sense to simply add the @qml.qfunc_transform here, and just call this add_noise:

@qml.qfunc_transform def add_noise(tape, noisy_op, ...): return new_tape

Then, both approaches are accessible:

@qml.qnode(dev) @add_noise(qml.AmplitudeDamping, 0.2, position="end") def f(w, x, y, z): qml.RX(w, wires=0) qml.RY(x, wires=1) qml.CNOT(wires=[0, 1]) qml.RY(y, wires=0) qml.RX(z, wires=1) return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1)) >>> print(qml.draw(f)(0.9, 0.4, 0.5, 0.6)) 0: ──RX(0.9)──╭C──RY(0.5)──AmplitudeDamping(0.2)──╭┤ ⟨Z ⊗ Z⟩ 1: ──RY(0.4)──╰X──RX(0.6)──AmplitudeDamping(0.2)──╰┤ ⟨Z ⊗ Z⟩

The tape transform is accessible via transform.tape_fn:

with qml.tape.QuantumTape() as tape: qml.RX(0.9, wires=0) qml.RY(0.4, wires=1) qml.CNOT(wires=[0, 1]) qml.RY(0.5, wires=0) qml.RX(0.6, wires=1) qml.expval(qml.PauliZ(0) @ qml.PauliZ(1)) new_tape = add_noise.tape_fn(tape, qml.AmplitudeDamping, 0.2, position="end") >>> print(new_tape.draw()) 0: ──RX(0.9)──╭C──RY(0.5)──AmplitudeDamping(0.2)──╭┤ ⟨Z ⊗ Z⟩ 1: ──RY(0.4)──╰X──RX(0.6)──AmplitudeDamping(0.2)──╰┤ ⟨Z ⊗ Z⟩

This solves the docstring issue, and avoids the duplication.

Alternatively, with a two-line change to qfunc_transform, we can support single-dispatch-style behaviour, where the transform acts correctly depending on the argument type:

new_qnode = add_noise(qml.AmplitudeDamping, 0.2, position="end")(qnode) new_tape = add_noise(qml.AmplitudeDamping, 0.2, position="end")(tape)

This is supported by the following diff:

diff --git a/pennylane/transforms/qfunc_transforms.py b/pennylane/transforms/qfunc_transforms.py index b631f39c..6394c0d1 100644 --- a/pennylane/transforms/qfunc_transforms.py +++ b/pennylane/transforms/qfunc_transforms.py @@ -172,6 +172,9 @@ class single_tape_transform: def _create_qfunc_internal_wrapper(fn, tape_transform, transform_args, transform_kwargs): """Convenience function to create the internal wrapper function generated by the qfunc_transform decorator""" + if isinstance(fn, qml.tape.QuantumTape): + return tape_transform(fn, *transform_args, **transform_kwargs) + if not callable(fn): raise ValueError( f"The qfunc to transform, {fn}, does not appear "

if you would like to add it.

Thanks Josh for the feedback! We now have add_noise which primarily acts as a qfunc/QNode transform, with the tape transform accessed via add_noise.tape_fn. I briefly tried the single dispatch approach suggested in your last comment but had some trouble, so thought it best to leave out for now.

Only downside is that the docstrings issue is still there:

and

Might be livable-with though.

Co-authored-by: Josh Izaac <josh146@gmail.com>

* Move device transform API into the @qfunc_transform decorator * linting * add to changelog * support drawing expansions * typo * Apply suggestions from code review Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * Update pennylane/transforms/draw.py Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com>

trbromley · 2021-10-27T12:52:26Z

Thanks @josh146 for the help with #1809! This PR is now ready for a final look through. Am really happy we converged on a single-dispatch approach.

doc/releases/changelog-dev.md

glassnotes

Just left a couple minor suggestions, looks awesome @trbromley ! 🎉

glassnotes · 2021-10-27T13:48:13Z

pennylane/transforms/draw.py

+              the internal circuit such that all circuit operations are supported by the gradient
+              method.
+
+            - ``device``: The QNode will attempt to decompose the internal circuit


Ohhh this will be really really convenient for future device-specific compilation transforms like allocation and routing, it will easily allow you to see the before and after!

glassnotes · 2021-10-27T13:53:56Z

pennylane/transforms/insert_ops.py

+
+@qfunc_transform
+def insert(
+    circuit: Union[callable, QuantumTape, Device],


Nothing actionable for this suggestion, but I did not notice there were type hints during my first review. I think this is the first time I've seen them in PL; is it something we'd like to add to more stuff, or is it just for convenience here because this transform can be called in a lot of different ways?

The official policy so far has been that type hinting is optional and up to the developer; feel free to add it if you would like, but it is not essential.

Perhaps something we will evaluate down the line, especially if we decide to start using mypy on the CI.

I really like type hinting because my IDE lets me know if I'm returning/passing arguments different to the type hints. Though the rendering in the docstrings doesn't look great.

pennylane/transforms/insert_ops.py

glassnotes · 2021-10-27T13:55:34Z

pennylane/transforms/insert_ops.py

+            acting on a single qubit, to be inserted into the circuit
+        op_args (tuple or float): the arguments fed to the operation, either as a tuple or a single
+            float
+        position (str): Specification of where to add the operation. Should be one of: ``"all"`` to


Makes sense! 💯

glassnotes · 2021-10-27T13:59:35Z

tests/transforms/test_insert_ops.py

+            qml.RX(0.6, wires=1)
+            qml.expval(qml.PauliZ(0) @ qml.PauliZ(1))
+
+        assert all(o1.name == o2.name for o1, o2 in zip(tape.operations, tape_exp.operations))


Yes no worries, and agree, a broader equivalence check mechanism would be really handy!

glassnotes · 2021-10-27T14:04:32Z

pennylane/transforms/insert_ops.py

+            add the operation after all gates; ``"start"`` to add the operation to all wires
+            at the start of the circuit; ``"end"`` to add the operation to all wires at the


Perhaps mention here the interaction with state prep operations and 'start' option. (Or maybe better in usage details?)

Good point, made that change: 96cb3d0

Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Olivia Di Matteo <2068515+glassnotes@users.noreply.github.com>

trbromley

Thanks @josh146 @glassnotes for the reviews!

trbromley · 2021-10-27T14:17:29Z

pennylane/transforms/insert_ops.py

+            add the operation after all gates; ``"start"`` to add the operation to all wires
+            at the start of the circuit; ``"end"`` to add the operation to all wires at the


Good point, made that change: 96cb3d0

trbromley added 15 commits October 22, 2021 08:15

Add to init

c470e10

Add to docs

76eb9c8

Add base transform

68494fd

Add tests

801496c

Add qfunc transform

36212a4

Add qfunc transform

3e5057a

Add test

8575a59

Add state prep support

c0ad944

Add tests

887f878

Add to changelog

e94095a

Merge branch 'master' into add_noisy_tape_transform

e50402c

Update changelog

f738e39

Fix changelog

8e2ef87

Fix codefactor

73dbc9f

Fix CI

89feb7a

trbromley added 2 commits October 22, 2021 16:04

Update docstrings

9433575

Fix

1ade5b3

trbromley commented Oct 22, 2021

View reviewed changes

trbromley added 11 commits October 25, 2021 08:32

Rename

a4d64f3

Add to init

2f1bbd8

Improve docstrings

5940470

Add to changelog

807b33a

Add to docstring

31e5b2c

Add test for add_noise_to_dev

7aed33c

Fix CI

59b6f9e

Attempt

d06d66f

Revert qfunc changes

de8594d

Update tests

71a7075

Fix CI

7406919

trbromley and others added 8 commits October 26, 2021 20:22

Remove single_tape_transform

a3b3248

Update pennylane/transforms/insert_ops.py

4f71fb9

Co-authored-by: Josh Izaac <josh146@gmail.com>

Update pennylane/transforms/insert_ops.py

dda77e3

Co-authored-by: Josh Izaac <josh146@gmail.com>

Update pennylane/transforms/insert_ops.py

5dead27

Co-authored-by: Josh Izaac <josh146@gmail.com>

Update

171f6ef

Add single dispatch

eca476b

Update changelog

540f253

Work on docstrings

94c69db

josh146 mentioned this pull request Oct 27, 2021

Move device transform API into the @qfunc_transform decorator #1809

Merged

trbromley and others added 6 commits October 27, 2021 07:35

Merge branch 'master' into add_noisy_tape_transform

42ea5bd

Add to changelog

c3cca25

Add to tests

e23ff95

Update docstrings

503f2d2

Update docstrings

788a68b

trbromley requested a review from josh146 October 27, 2021 12:52

josh146 approved these changes Oct 27, 2021

View reviewed changes

doc/releases/changelog-dev.md Outdated Show resolved Hide resolved

glassnotes approved these changes Oct 27, 2021

View reviewed changes

trbromley and others added 3 commits October 27, 2021 10:10

Merge branch 'master' into add_noisy_tape_transform

c76abaa

Apply suggestions from code review

62c58b5

Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Olivia Di Matteo <2068515+glassnotes@users.noreply.github.com>

Mention state preps

96cb3d0

trbromley commented Oct 27, 2021

View reviewed changes

trbromley merged commit 37d4254 into master Oct 27, 2021

trbromley deleted the add_noisy_tape_transform branch October 27, 2021 14:35

trbromley mentioned this pull request Oct 29, 2021

Add error mitigation batch transform #1813

Merged

This was referenced Nov 2, 2021

Update Device.expand_fn #1840

Merged

QVM runs with latest PennyLane PennyLaneAI/pennylane-rigetti#87

Closed

This was referenced Nov 8, 2021

Insert only for certain gates when using qml.transforms.insert #1876

Closed

Allow inserting multiqubit operations in qml.transforms.insert #1877

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add transform to insert operations into circuits #1795

Add transform to insert operations into circuits #1795

trbromley commented Oct 22, 2021 •

edited

Loading

codecov bot commented Oct 22, 2021 •

edited

Loading

trbromley Oct 22, 2021

trbromley Oct 22, 2021

josh146 Oct 25, 2021

josh146 Oct 25, 2021

josh146 Oct 25, 2021 •

edited

Loading

trbromley Oct 25, 2021

trbromley Oct 25, 2021

trbromley commented Oct 27, 2021

glassnotes left a comment

glassnotes Oct 27, 2021

glassnotes Oct 27, 2021

josh146 Oct 27, 2021

trbromley Oct 27, 2021

glassnotes Oct 27, 2021

glassnotes Oct 27, 2021

glassnotes Oct 27, 2021

trbromley Oct 27, 2021

trbromley left a comment

trbromley Oct 27, 2021

		add the operation after all gates; ``"start"`` to add the operation to all wires
		at the start of the circuit; ``"end"`` to add the operation to all wires at the

Add transform to insert operations into circuits #1795

Add transform to insert operations into circuits #1795

Conversation

trbromley commented Oct 22, 2021 • edited Loading

codecov bot commented Oct 22, 2021 • edited Loading

Codecov Report

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 25, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

trbromley commented Oct 27, 2021

glassnotes 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

Choose a reason for hiding this comment

trbromley left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

trbromley commented Oct 22, 2021 •

edited

Loading

codecov bot commented Oct 22, 2021 •

edited

Loading

josh146 Oct 25, 2021 •

edited

Loading