[unitaryHACK] Add reset error to noise channels #1321

nahumsa · 2021-05-15T19:40:47Z

Context: Add Reset Error to noise channels

Description of the Change: Add Single-qubit Reset error channel to pennylane.

Benefits: Add a new noise model that simulates the reset of qubits to 0 or 1 given a probability.

Possible Drawbacks:

Related GitHub Issues: #971

josh146 · 2021-05-17T06:48:26Z

Hi @nahumsa! Thanks for this PR -- this is a great issue to tackle. Just a bit of housekeeping before we can provide a review:

The formatting check is failing. All that is required is that the PennyLane source code and tests folder is 'blacked' using the Black formatting tool:
```
pip install black
black -l 100 pennylane
black -l 100 tests
```
The documentation check is also failing, due to a missing space after the backtick in the docstring:
```
:math:`p_1 \in [0, 1]`is
```
You can click on the 'details' link beside the docs/readthedocs.org:pennylane CI check to view what the rendered documentation for your PR looks like, to make sure it looks okay when built :)
Finally, it looks like there is an issue with the new tests, causing the test CI to fail.

Let us know when these have been addressed, and the PR is ready for review! Alternatively, feel free to ask any questions.

…into add-reset-error

nahumsa · 2021-05-17T11:29:46Z

Thanks for the response, @josh146 ! I will fix those issues and let you know when it's ready for a review.

Changed the probabilities of tests, because the sum of probs was greater than 1, and fix the gradient circuit parameters

nahumsa · 2021-05-17T14:39:14Z

I think that this is ready for review.

josh146

Thanks for this great PR @nahumsa! In particular, fantastic tests.

However, it doesn't appear that the tests are passing; they are giving the error

AssertionError: Gradient recipe must have one entry for each parameter!

The grad_recipe you have defined provides a gradient recipe only for the first parameter:

grad_recipe = ([[1, 0, 1], [-1, 0, 0]],)

You will need to also provide the gradient recipe for the second parameter, p_1. Let me know if you have any questions!

.github/CHANGELOG.md

josh146 · 2021-05-17T15:53:35Z

pennylane/ops/channel.py

+    @classmethod
+    def _kraus_matrices(cls, *params):
+        p_0, p_1 = params[0], params[1]
+        K0 = np.sqrt(1 - p_0 - p_1) * np.eye(2)
+        K1 = np.sqrt(p_0) * np.array([[1, 0], [0, 0]])
+        K2 = np.sqrt(p_1) * np.array([[0, 0], [0, 1]])
+        return [K0, K1, K2]


josh146 · 2021-05-17T15:54:58Z

tests/ops/test_channel_ops.py

+    @pytest.mark.parametrize("angle", np.linspace(0, 2 * np.pi, 7))
+    def test_grad_reset(self, angle, tol):
+        """Test that analytical gradient is computed correctly for different states. Channel
+        grad recipes are independent of channel parameter"""
+
+        dev = qml.device("default.mixed", wires=1)
+        p_0, p_1 = 0.5, 0.5
+
+        @qml.qnode(dev)
+        def circuit(p_0, p_1):
+            qml.RX(angle, wires=0)
+            qml.ResetError(p_0, p_1, wires=0)
+            return qml.expval(qml.PauliZ(0))
+
+        gradient = np.squeeze(qml.grad(circuit)(p_0, p_1))
+        assert gradient == circuit(1) - circuit(0)
+        assert np.allclose(gradient, (-2 * np.cos(angle)))


Great test @nahumsa! Did you calculate the expected gradient value, -2 cos(angle), by hand?

Yeah, I'm not sure about the analytical gradients for this one, I was using TestBitFlip as a reference.
Maybe could you clarify a bit more about the gradient of this channel?

One of the best sources for learning more about the analytic gradients of noisy channels is https://johannesjakobmeyer.com/blog/004-noisy-parameter-shift/, or alternatively, the PennyLane demonstration: https://pennylane.ai/qml/demos/tutorial_noisy_circuit_optimization.html

In this particular case, I think @ixfoduap calculated the analytic value by hand, so hopefully he will be able to shed some light!

Thanks for the followup. I'll read those references.

Hi @nahumsa, this is a great PR!

I'm not sure about the analytical gradients. The way we calculate analytical gradients is using this formula, which also appears in the blog post that Josh linked to:

It is valid when the channel can be expressed as a linear combination of channels. This typically means that the Kraus matrices themselves should be a valid channel, which in practice often boils down to the Kraus matrices being unitaries.

In your case, the projectors |0><0| and |1><1| are not by themselves valid channels, so it's not accurate that the reset channel can be expressed as a linear combination of channels. This means that the analytical formula should not apply here.

I believe that the reason the tests are passing is because you are setting both p_0 and p_1 equal to each other, which effectively makes this the identity channel --> sqrt(p) |0><0| + sqrt(p) |1><1| = sqrt(p) (|0><0|+|1><1|) = sqrt(p) \identity and we end up with the gradient of the gate in the circuit.

I would expect the test to fail if you make p_0 != p_1 and compute the gradient by hand there. It's worth checking that this is the case. In any case, the gradient test should be checking gradients for many values of p_0 and p_1.

If the analytical gradients do not apply here, it is not a big deal; it just means you should change the gradient method to "F" (finite difference) like it is done for other channels.

Hi @ixfoduap! Thanks for the clarification. After reading the reference, I think that it is not possible to get analytic gradient by the formulas used on other channels.
Thanks for the catch about the tests.

Thanks @ixfoduap!

@nahumsa let us know when you have the tests passing, or if you need a hand anywhere :)

@josh146 and @ixfoduap, after revising some math, I found that I missed two Kraus operators that constitutes the ResetError channel! Sorry about that, I always find quantum channels quite tricky 😅

josh146 · 2021-05-17T16:13:33Z

Oh, and don't forget to add your name to the Contributors section of the changelog!

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

nahumsa · 2021-05-17T16:17:59Z

Oh, and don't forget to add your name to the Contributors section of the changelog!

Sure! :)

ixfoduap · 2021-05-18T19:03:10Z

tests/ops/test_channel_ops.py

+    @pytest.mark.parametrize("angle", np.linspace(0, 2 * np.pi, 7))
+    def test_grad_reset(self, angle, tol):
+        """Test that analytical gradient is computed correctly for different states. Channel
+        grad recipes are independent of channel parameter"""
+
+        dev = qml.device("default.mixed", wires=1)
+        p_0, p_1 = 0.5, 0.5
+
+        @qml.qnode(dev)
+        def circuit(p_0, p_1):
+            qml.RX(angle, wires=0)
+            qml.ResetError(p_0, p_1, wires=0)
+            return qml.expval(qml.PauliZ(0))
+
+        gradient = np.squeeze(qml.grad(circuit)(p_0, p_1))
+        assert gradient == circuit(1) - circuit(0)
+        assert np.allclose(gradient, (-2 * np.cos(angle)))


Hi @nahumsa, this is a great PR!

I'm not sure about the analytical gradients. The way we calculate analytical gradients is using this formula, which also appears in the blog post that Josh linked to:

It is valid when the channel can be expressed as a linear combination of channels. This typically means that the Kraus matrices themselves should be a valid channel, which in practice often boils down to the Kraus matrices being unitaries.

In your case, the projectors |0><0| and |1><1| are not by themselves valid channels, so it's not accurate that the reset channel can be expressed as a linear combination of channels. This means that the analytical formula should not apply here.

I believe that the reason the tests are passing is because you are setting both p_0 and p_1 equal to each other, which effectively makes this the identity channel --> sqrt(p) |0><0| + sqrt(p) |1><1| = sqrt(p) (|0><0|+|1><1|) = sqrt(p) \identity and we end up with the gradient of the gate in the circuit.

I would expect the test to fail if you make p_0 != p_1 and compute the gradient by hand there. It's worth checking that this is the case. In any case, the gradient test should be checking gradients for many values of p_0 and p_1.

If the analytical gradients do not apply here, it is not a big deal; it just means you should change the gradient method to "F" (finite difference) like it is done for other channels.

…into add-reset-error

codecov · 2021-05-19T12:07:04Z

Codecov Report

Merging #1321 (b35a7be) into master (98a7876) will decrease coverage by 0.94%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1321      +/-   ##
==========================================
- Coverage   99.10%   98.16%   -0.95%     
==========================================
  Files         141      145       +4     
  Lines       10575    11093     +518     
==========================================
+ Hits        10480    10889     +409     
- Misses         95      204     +109

Impacted Files	Coverage Δ
pennylane/ops/channel.py	`100.00% <100.00%> (ø)`
pennylane/tape/operation_recorder.py	`64.51% <0.00%> (-35.49%)`	⬇️
pennylane/measure.py	`91.76% <0.00%> (-8.24%)`	⬇️
pennylane/circuit_drawer/grid.py	`88.70% <0.00%> (-7.72%)`	⬇️
pennylane/grouping/optimize_measurements.py	`92.30% <0.00%> (-7.70%)`	⬇️
pennylane/vqe/vqe.py	`92.97% <0.00%> (-7.03%)`	⬇️
pennylane/configuration.py	`93.10% <0.00%> (-6.90%)`	⬇️
pennylane/qnn/torch.py	`95.08% <0.00%> (-4.92%)`	⬇️
pennylane/qnn/keras.py	`95.38% <0.00%> (-4.62%)`	⬇️
pennylane/about.py	`95.45% <0.00%> (-4.55%)`	⬇️
... and 24 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 98a7876...b35a7be. Read the comment docs.

josh146

Great contribution @nahumsa, looks ready to merge on my end! I might wait to see if @ixfoduap has any requests.

ixfoduap · 2021-05-19T17:44:32Z

pennylane/ops/channel.py

+                \end{bmatrix}
+
+    .. math::
+        K_2 = \sqrt{p_0}\begin{bmatrix}


@nahumsa Why did you change the definition of the channel?
Before approving, could you please point me to a reference where this channel is defined? I would like to make sure that the name "reset error" coincides with the standard usage. Thanks!

Hi @ixfoduap ! I was reading through the qiskit's docs and found a direct reference about the ResetError Kraus operators:

This was the main reason that I changed the definition of the channel, do you think that this is reasonable?

Perfect! Thanks for the clarification!

ixfoduap

nahumsa added 4 commits May 15, 2021 16:04

add ResetError channel

123ddda

updated test parameters

87c419d

updated docs

88e9375

add entry to changelog

3ccbbac

nahumsa changed the title ~~[WIP] [unitaryhack] Add reset error to noise channels~~ [unitaryhack] Add reset error to noise channels May 15, 2021

Merge branch 'master' into add-reset-error

bb468e8

josh146 added the unitaryhack Dedicated issue for Unitary Fund open-source hackathon label May 17, 2021

josh146 requested a review from ixfoduap May 17, 2021 06:43

nahumsa added 3 commits May 17, 2021 08:28

run black

44ae9cd

Merge branch 'add-reset-error' of https://github.com/nahumsa/pennylane …

891e3c8

…into add-reset-error

fix docs

53b6986

nahumsa added 2 commits May 17, 2021 08:34

fix tests

c0a1256

Changed the probabilities of tests, because the sum of probs was greater than 1, and fix the gradient circuit parameters

fix name of tests

b1e01de

Merge branch 'master' into add-reset-error

b1bc4f5

josh146 requested changes May 17, 2021

View reviewed changes

Update changelog

85b7bb5

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

nahumsa and others added 3 commits May 17, 2021 13:47

add name to contributors

c384377

add grad recipe for p_1

f020b48

Merge branch 'master' into add-reset-error

7e66d67

ixfoduap requested changes May 18, 2021

View reviewed changes

nahumsa and others added 5 commits May 18, 2021 17:09

change grad_method to "F"

31d91e1

add ResetError when testing if it is trace preserving

af53e0d

remove analytic gradient test for ResetError

4ce0c52

Merge branch 'master' into add-reset-error

2783fc5

Merge branch 'master' into add-reset-error

05c5f38

nahumsa added 4 commits May 19, 2021 08:38

updated kraus operators

61ed7d8

Merge branch 'add-reset-error' of https://github.com/nahumsa/pennylane …

2623fd3

…into add-reset-error

fix channel names

e8d16b1

fix test channel names

77ab397

nahumsa changed the title ~~[unitaryhack] Add reset error to noise channels~~ [unitaryHACK] Add reset error to noise channels May 19, 2021

josh146 approved these changes May 19, 2021

View reviewed changes

josh146 requested a review from ixfoduap May 19, 2021 14:57

ixfoduap reviewed May 19, 2021

View reviewed changes

ixfoduap approved these changes May 19, 2021

View reviewed changes

Merge branch 'master' into add-reset-error

b35a7be

josh146 merged commit 24e9360 into PennyLaneAI:master May 20, 2021

This was referenced Aug 16, 2021

Create qml.PauliError #1529

Closed

Add support for qml.ResetError in default.mixed #1530

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[unitaryHACK] Add reset error to noise channels #1321

[unitaryHACK] Add reset error to noise channels #1321

nahumsa commented May 15, 2021 •

edited

Loading

josh146 commented May 17, 2021

nahumsa commented May 17, 2021

nahumsa commented May 17, 2021

josh146 left a comment

josh146 May 17, 2021

josh146 May 17, 2021

nahumsa May 17, 2021

josh146 May 18, 2021

josh146 May 18, 2021

nahumsa May 18, 2021 •

edited

Loading

ixfoduap May 18, 2021

nahumsa May 18, 2021 •

edited

Loading

josh146 May 19, 2021

josh146 May 19, 2021

nahumsa May 19, 2021

josh146 commented May 17, 2021

nahumsa commented May 17, 2021

ixfoduap May 18, 2021

codecov bot commented May 19, 2021 •

edited

Loading

josh146 left a comment

ixfoduap May 19, 2021

nahumsa May 19, 2021

ixfoduap May 19, 2021

ixfoduap left a comment

[unitaryHACK] Add reset error to noise channels #1321

[unitaryHACK] Add reset error to noise channels #1321

Conversation

nahumsa commented May 15, 2021 • edited Loading

josh146 commented May 17, 2021

nahumsa commented May 17, 2021

nahumsa commented May 17, 2021

josh146 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

nahumsa May 18, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

nahumsa May 18, 2021 • edited Loading

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 commented May 17, 2021

nahumsa commented May 17, 2021

Choose a reason for hiding this comment

codecov bot commented May 19, 2021 • edited Loading

Codecov Report

josh146 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

ixfoduap left a comment

Choose a reason for hiding this comment

nahumsa commented May 15, 2021 •

edited

Loading

nahumsa May 18, 2021 •

edited

Loading

nahumsa May 18, 2021 •

edited

Loading

codecov bot commented May 19, 2021 •

edited

Loading