add LinearAncillaComposite #530

pau557 · 2024-08-15T21:14:13Z

In scenarios where the h biases are not available or sufficient in range, one can use auxiliary qubits polarized with a large flux bias and coupled to the data qubit. This method:

Requires that each qubit is coupled to a non-programmed qubit
But, ancilla qubits can be shared between data qubits
The magnitude of the bias is limited by the magnitude of the coupler, but in the case of asymetric coupling range, one can use the strongest coupling available and control the bias sign with the flux bias sign

This PR adds a LinearAncillaComposite that implements this technique.

For the reviewers, please take into account future extensions of this method:

One can use multiple ancilla qubits to extend the h-range and avoid shrinking energy scales of the problem. This is relevant for QUBO --> Ising translation.
While the current implementation may work on embedded problems as is, we need an embedding-aware version that relaxes the condition of each qubit being coupled to an ancilla to each chain being coupled to at least one ancilla.

Please educate me on how to use the correct Sphynx syntax

codecov · 2024-08-15T21:20:23Z

Codecov Report

Attention: Patch coverage is 83.33333% with 12 lines in your changes missing coverage. Please review.

Project coverage is 86.57%. Comparing base (b233941) to head (5867eba).
Report is 31 commits behind head on master.

Files with missing lines	Patch %	Lines
dwave/system/composites/linear_ancilla.py	83.09%	12 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #530      +/-   ##
==========================================
- Coverage   89.65%   86.57%   -3.09%     
==========================================
  Files          24       25       +1     
  Lines        1760     1832      +72     
==========================================
+ Hits         1578     1586       +8     
- Misses        182      246      +64

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚨 Try these New Features:

Flaky Tests Detection - Detect and resolve failed and flaky tests

JoelPasvolsky

This can be very useful, @pau557. I made a few comments inline.

Needs to be added to docs here: https://github.com/dwavesystems/dwave-system/blob/master/docs/reference/composites.rst

dwave/system/composites/linear_ancilla.py

releasenotes/notes/add-linear-ancilla-composite-3281ed6733b0f0c7.yaml

dwave/system/composites/linear_ancilla.py

tests/test_linear_ancilla_composite.py

dwave/system/composites/linear_ancilla.py

tests/test_linear_ancilla_composite.py

JoelPasvolsky

Thank you

docs/reference/composites.rst

pau557 · 2024-11-19T00:19:39Z

Any more comments or ready to merge?

JoelPasvolsky · 2024-11-19T15:03:36Z

Any more comments or ready to merge?

@pau557 there are comments I think you might have missed, especially ones that use GitHub Suggested Change that have not been added. Examples: 1, 2, 3, 4, 5, etc. Can you look through the comments again?

pau557 · 2024-11-22T02:41:04Z

I had missed many comments, apologies @JoelPasvolsky. I think everything is addressed now

JoelPasvolsky

Just a few small comments.

dwave/system/composites/linear_ancilla.py

randomir

LGTM

randomir · 2024-11-26T16:25:13Z

dwave/system/composites/linear_ancilla.py

+        bqm: dimod.BinaryQuadraticModel,
+        *,
+        h_tolerance: numbers.Number = 0,
+        default_flux_bias_range: Sequence[float] = (-0.005, 0.005),


why not be more precise with:

Suggested change

default_flux_bias_range: Sequence[float] = (-0.005, 0.005),

default_flux_bias_range: tuple[float, float] = (-0.005, 0.005),

CircleCI didn't like that on python 3.8...

File "/home/circleci/project/dwave/system/composites/linear_ancilla.py", line 96, in LinearAncillaComposite default_flux_bias_range: tuple[float, float] = (-0.005, 0.005), TypeError: 'type' object is not subscriptable

Yup. But we dropped support for 3.8 in the main branch since you opened this PR.. So, best to rebase to master.

@pau557, would you prefer to rebase, or should I do it?

dwave/system/composites/linear_ancilla.py

randomir · 2024-11-26T20:54:01Z

@pau557, I think we can merge this when docs comments are addressed.

pau557 · 2024-11-27T02:50:13Z

How do I tell CircleCI not to run this example with a mock sampler?

/home/circleci/project/dwave/system/testing.py:325: UserWarning: 'flux_biases' parameter is valid for DWaveSampler(), but not mocked in MockDWaveSampler().
  warnings.warn(f'{kw!r} parameter is valid for DWaveSampler(), '
/home/circleci/project/dwave/system/testing.py:325: UserWarning: 'fast_anneal' parameter is valid for DWaveSampler(), but not mocked in MockDWaveSampler().
  warnings.warn(f'{kw!r} parameter is valid for DWaveSampler(), '
WARNING: **********************************************************************
File "../dwave/system/composites/linear_ancilla.py", line ?, in default
Failed example:
    sampleset.first.energy
Expected:
    -3
Got:
    1.0

jackraymond · 2024-11-27T04:08:05Z

How do I tell CircleCI not to run this example with a mock sampler?

/home/circleci/project/dwave/system/testing.py:325: UserWarning: 'flux_biases' parameter is valid for DWaveSampler(), but not mocked in MockDWaveSampler().
  warnings.warn(f'{kw!r} parameter is valid for DWaveSampler(), '
/home/circleci/project/dwave/system/testing.py:325: UserWarning: 'fast_anneal' parameter is valid for DWaveSampler(), but not mocked in MockDWaveSampler().
  warnings.warn(f'{kw!r} parameter is valid for DWaveSampler(), '
WARNING: **********************************************************************
File "../dwave/system/composites/linear_ancilla.py", line ?, in default
Failed example:
    sampleset.first.energy
Expected:
    -3
Got:
    1.0

You can mock the flux_biases parameter with h if you want this to behave correctly.
You can also suppress the warnings. e.g.
https://github.com/AndyZzzZzzZzz/shimming-tutorial/blob/development/tutorial_code/helpers/sampler_wrapper.py
I think its worth testing

dwave/system/composites/linear_ancilla.py

jackraymond

Couple of suggestions, these needn't be blockers to integration.

jackraymond · 2024-11-27T17:19:19Z

dwave/system/composites/linear_ancilla.py

+        flux_biases = [0] * qpu_properties["num_qubits"]
+
+        _bqm = bqm.copy()
+        used_ancillas = defaultdict(list)


In practice we may want to share ancillas. This might allow some problems to be programmed that are impossible with unique ancilla per variable. Might be worth adding a comment to this effect somewhere, with a view to a future feature expansion (probably an optional argument).

We share ancillas. We're just biased towards picking the least used ones

https://github.com/dwavesystems/dwave-system/pull/530/files#diff-558dca34f40df626e648cc5dc9cb3f09dcafe585a4e5860a540228d64b33a847R171

jackraymond · 2024-11-27T19:19:19Z

tests/test_linear_ancilla_composite.py

+import networkx as nx
+import numpy as np
+
+from dwave.system.testing import MockDWaveSampler


The MockSampler can be amended so as to emulate flux_biases as Ising model linear biases:
https://github.com/AndyZzzZzzZzz/shimming-tutorial/blob/andy/tutorial_code/helpers/sampler_wrapper.py

I recommend you use a wrapper like this one and then test optimization using MockSampler().sample

The tests in test_linear_ancilla_composite do not check for sample correctness. Just for the implementation of ancillas and flux biases in the call. So the MockSampler works for it.

The error that I posted yesterday was related to circle CI testing the example in a docstring. In that case I am not sure it's possible to wrap with this fancier mock sampler. If it's not trivial to do so I would skip

JoelPasvolsky · 2024-11-29T19:18:09Z

Looks like an incorrect rebase: did you miss pulling the latest master branch?

pau557 · 2024-11-29T19:55:59Z

Looks like an incorrect rebase: did you miss pulling the latest master branch?

Yes, I did it wrong. @randomir is helping with it.

randomir · 2024-11-29T21:04:26Z

Because this PR was opened from a fork's master branch, rebasing would be messy, i.e. history would be pretty much lost.

In order to keep the conversation here relevant (and retain @pau557's authorship of commits), I have rebased in a new branch and verified all tests pass in a separate draft PR, #546. So we can merge this as-is.

add LinearAncillaComposite

acc5e68

JoelPasvolsky reviewed Aug 16, 2024

View reviewed changes

pau557 added 2 commits August 16, 2024 16:06

formatting

b7d11f5

add note abou auto_scale support

983e5ce

arcondello reviewed Aug 29, 2024

View reviewed changes

Add LinearAncillaComposite to docs

d8416ff

JoelPasvolsky reviewed Oct 10, 2024

View reviewed changes

docs/reference/composites.rst Outdated Show resolved Hide resolved

docs/reference/composites.rst Outdated Show resolved Hide resolved

pau557 added 2 commits October 10, 2024 16:53

minor formatting

74f4e31

change description from generalized to specific

63e4026

docstring formatting, variable naming

5867eba

JoelPasvolsky reviewed Nov 22, 2024

View reviewed changes

dwave/system/composites/linear_ancilla.py Outdated Show resolved Hide resolved

dwave/system/composites/linear_ancilla.py Outdated Show resolved Hide resolved

dwave/system/composites/linear_ancilla.py Show resolved Hide resolved

randomir approved these changes Nov 26, 2024

View reviewed changes

add example to composite docstring; formatting

6751401

JoelPasvolsky reviewed Nov 27, 2024

View reviewed changes

dwave/system/composites/linear_ancilla.py Outdated Show resolved Hide resolved

dwave/system/composites/linear_ancilla.py Outdated Show resolved Hide resolved

randomir reviewed Nov 27, 2024

View reviewed changes

dwave/system/composites/linear_ancilla.py Outdated Show resolved Hide resolved

skip docs test; minor formatting

080bfa2

jackraymond approved these changes Nov 27, 2024

View reviewed changes

JoelPasvolsky approved these changes Nov 27, 2024

View reviewed changes

randomir force-pushed the master branch from 46a73d5 to 080bfa2 Compare November 29, 2024 20:53

randomir mentioned this pull request Nov 29, 2024

#530 rebased #546

Closed

randomir merged commit 5d64d83 into dwavesystems:master Nov 29, 2024
13 of 16 checks passed

randomir mentioned this pull request Dec 2, 2024

Release 1.28.0 #548

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add LinearAncillaComposite #530

add LinearAncillaComposite #530

pau557 commented Aug 15, 2024

codecov bot commented Aug 15, 2024 •

edited

Loading

JoelPasvolsky left a comment

JoelPasvolsky left a comment

pau557 commented Nov 19, 2024

JoelPasvolsky commented Nov 19, 2024

pau557 commented Nov 22, 2024 •

edited

Loading

JoelPasvolsky left a comment

randomir left a comment

randomir Nov 26, 2024

pau557 Nov 27, 2024 •

edited

Loading

randomir Nov 27, 2024

randomir Nov 28, 2024

randomir commented Nov 26, 2024

pau557 commented Nov 27, 2024 •

edited

Loading

jackraymond commented Nov 27, 2024 •

edited

Loading

jackraymond left a comment

jackraymond Nov 27, 2024

pau557 Nov 27, 2024

jackraymond Nov 27, 2024

pau557 Nov 27, 2024

JoelPasvolsky commented Nov 29, 2024 •

edited

Loading

pau557 commented Nov 29, 2024

randomir commented Nov 29, 2024

	default_flux_bias_range: Sequence[float] = (-0.005, 0.005),
	default_flux_bias_range: tuple[float, float] = (-0.005, 0.005),

add LinearAncillaComposite #530

add LinearAncillaComposite #530

Conversation

pau557 commented Aug 15, 2024

codecov bot commented Aug 15, 2024 • edited Loading

Codecov Report

JoelPasvolsky left a comment

Choose a reason for hiding this comment

JoelPasvolsky left a comment

Choose a reason for hiding this comment

pau557 commented Nov 19, 2024

JoelPasvolsky commented Nov 19, 2024

pau557 commented Nov 22, 2024 • edited Loading

JoelPasvolsky left a comment

Choose a reason for hiding this comment

randomir left a comment

Choose a reason for hiding this comment

randomir Nov 26, 2024

Choose a reason for hiding this comment

pau557 Nov 27, 2024 • edited Loading

Choose a reason for hiding this comment

randomir Nov 27, 2024

Choose a reason for hiding this comment

randomir Nov 28, 2024

Choose a reason for hiding this comment

randomir commented Nov 26, 2024

pau557 commented Nov 27, 2024 • edited Loading

jackraymond commented Nov 27, 2024 • edited Loading

jackraymond left a comment

Choose a reason for hiding this comment

jackraymond Nov 27, 2024

Choose a reason for hiding this comment

pau557 Nov 27, 2024

Choose a reason for hiding this comment

jackraymond Nov 27, 2024

Choose a reason for hiding this comment

pau557 Nov 27, 2024

Choose a reason for hiding this comment

JoelPasvolsky commented Nov 29, 2024 • edited Loading

pau557 commented Nov 29, 2024

randomir commented Nov 29, 2024

codecov bot commented Aug 15, 2024 •

edited

Loading

pau557 commented Nov 22, 2024 •

edited

Loading

pau557 Nov 27, 2024 •

edited

Loading

pau557 commented Nov 27, 2024 •

edited

Loading

jackraymond commented Nov 27, 2024 •

edited

Loading

JoelPasvolsky commented Nov 29, 2024 •

edited

Loading