Bayesian Networks #5094

tonybruguier · 2022-03-17T12:51:50Z

This class implements Quantum Bayesian Networks as described in:
https://arxiv.org/abs/2004.14803

In addition, these write ups could be helpful:
https://towardsdatascience.com/create-a-quantum-bayesian-network-d26d7c5c4217
https://towardsdatascience.com/how-to-create-a-quantum-bayesian-network-5b011914b03e

tonybruguier · 2022-04-14T19:34:38Z

@MichaelBroughton
Do you have any comments for this PR? I think the overall approach was OK'ed during the cynq, but reconsidering is of course possible.

vtomole · 2022-04-14T19:40:09Z

The discussion from Cirq sync was about considering whether or not this should be in Cirq proper. I think someone was going to ask around internally to find out if anyone is interested in Bayesian Networks.

review-notebook-app · 2022-04-14T19:56:01Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

tonybruguier · 2022-04-14T20:03:08Z

@vtomole
My understanding is that the present PR should be in contrib and a Jupyter that uses this code to Gibbs sampling is of interest.

But I am totally OK with re-considering. I certainly don't want to pollute the repo and could abandon this PR if it is misguided, of course.

MichaelBroughton

Good first pass. I'm wondering if we've got the logic from the paper exactly captured. Are we handling the cases of the root vs child nodes ? Perhaps it might be worth recreating a test to match what is done in figure 10 for corectness sake ?

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py

MichaelBroughton · 2022-04-21T20:11:56Z

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py

+        for param, init_prob in self._init_probs:
+            yield self._generate_got_set_for_init_prob(qubit_map[param], init_prob)


Are we assuming all nodes are root nodes and none follow along with eqn 14 from the paper ?

Correct. We just specify the dependent nodes with a None probability. Also the user messes up and specify a node is independent (i.e. with an initial probability) and then assigns it as a dependent variable, the code will run fine (but return absurd results).

The reason I decided to do it this way, was that there are other ways to mess things up. We would have to check the graph is directed acyclic, but that seems like pulling a lot of graph code in. I can do it if you want, of course. For now, I just put a warning in the docstring.

Leaving unresolved. Let me know?

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py

MichaelBroughton · 2022-04-21T20:16:49Z

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py

+            if not isinstance(params, tuple):
+                raise ValueError('Conditional prob params must be a tuple.')


Nit: don't need to make type checking part of python APIs

Right, but I want to distinguish cases like
('q0')
versus
('q0',)

The first version is NOT correct, but without the check, the code will iterate over the characters 'q' and '0'. It's a very common mistake, so this could be a good help.

tonybruguier · 2022-04-21T22:30:42Z

Intermed

Good first pass. I'm wondering if we've got the logic from the paper exactly captured. Are we handling the cases of the root vs child nodes ? Perhaps it might be worth recreating a test to match what is done in figure 10 for corectness sake ?

I tried to recreate the network of figure 10, and I got this circuit:

It is simpler than figure 11 because of Gray optimization that I had added on top of the paper + a different definition of the gate. Also, note that I suspect the paper had the order of the qubits flipped.

Still, I tried to reproduce the values of table 1, and I was able to (except that the paper could have been more precise and write 0.4985 instead of 0.499)

tonybruguier · 2022-04-22T21:13:46Z

Thanks @MichaelBroughton . Now, the PR is ready for another iteration. I left a few comments unresolved to make sure I understood what you were asking. PTAL when you can?

tonybruguier · 2022-05-11T18:06:48Z

Friendly ping when you have the time? I understand that v1.0 takes priority, of course.

dabacon

@MichaelBroughton may want to take another look.

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py

tonybruguier · 2022-05-16T18:21:54Z

Thanks. PTAL.

CirqBot · 2022-06-08T18:12:29Z

Automerge cancelled: A status check is failing.

tonybruguier · 2022-06-08T19:07:55Z

@MichaelBroughton
I am not sure why automerge failed, but I re-pushed without any code change and it seems OK now.

CirqBot · 2022-06-22T18:07:10Z

Automerge cancelled: A status check is failing.

vtomole · 2022-06-22T18:11:12Z

@tonybruguier Can't merge due to some checks failing.

This class implements Quantum Bayesian Networks as described in: https://arxiv.org/abs/2004.14803 In addition, these write ups could be helpful: https://towardsdatascience.com/create-a-quantum-bayesian-network-d26d7c5c4217 https://towardsdatascience.com/how-to-create-a-quantum-bayesian-network-5b011914b03e

CirqBot added the size: L 250< lines changed <1000 label Mar 17, 2022

tonybruguier marked this pull request as ready for review March 17, 2022 16:15

tonybruguier requested review from a team, cduck and vtomole as code owners March 17, 2022 16:15

tonybruguier requested a review from maffoo March 17, 2022 16:15

MichaelBroughton self-assigned this Mar 17, 2022

tonybruguier requested review from ColemanCollins, HGSilveri, dabacon, gmauricio, ma5x, mrwojtek, nakardo, pschindler, verult and wcourtney as code owners April 14, 2022 19:55

tonybruguier force-pushed the cirq_qbn branch from ad5c52e to 54e7115 Compare April 14, 2022 19:58

tonybruguier force-pushed the cirq_qbn branch from 9938813 to 4b578b3 Compare April 19, 2022 04:01

MichaelBroughton reviewed Apr 21, 2022

View reviewed changes

dabacon approved these changes May 16, 2022

View reviewed changes

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py Outdated Show resolved Hide resolved

cirq-core/cirq/contrib/bayesian_network/bayesian_network_gate.py Outdated Show resolved Hide resolved

MichaelBroughton added the automerge Tells CirqBot to sync and merge this PR. (If it's running.) label Jun 8, 2022

MichaelBroughton added good for learning For beginners in QC, this will help picking up some knowledge. Bit harder than "good first issues" """"""""""""""""""" labels Jun 8, 2022

CirqBot added the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Jun 8, 2022

CirqBot removed automerge Tells CirqBot to sync and merge this PR. (If it's running.) front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. labels Jun 8, 2022

Quantum Bayesian Networks

00978a7

tonybruguier force-pushed the cirq_qbn branch from 9a9be6c to 00978a7 Compare June 8, 2022 18:49

maffoo removed the """"""""""""""""""" label Jun 15, 2022

tonybruguier and others added 3 commits June 22, 2022 08:55

Merge branch 'master' into cirq_qbn

131c084

Explicitely copy the state vector (merging changes)

20e3721

Merge branch 'master' into cirq_qbn

80b765e

vtomole added the automerge Tells CirqBot to sync and merge this PR. (If it's running.) label Jun 22, 2022

CirqBot added the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Jun 22, 2022

CirqBot removed automerge Tells CirqBot to sync and merge this PR. (If it's running.) front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. labels Jun 22, 2022

retrigger checks

5a04fd4

vtomole added the automerge Tells CirqBot to sync and merge this PR. (If it's running.) label Jun 22, 2022

CirqBot added the front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. label Jun 22, 2022

CirqBot merged commit b948a66 into quantumlib:master Jun 22, 2022

CirqBot removed automerge Tells CirqBot to sync and merge this PR. (If it's running.) front_of_queue_automerge CirqBot uses this label to indicate (and remember) what's being merged next. labels Jun 22, 2022

		for param, init_prob in self._init_probs:
		yield self._generate_got_set_for_init_prob(qubit_map[param], init_prob)

		if not isinstance(params, tuple):
		raise ValueError('Conditional prob params must be a tuple.')

Bayesian Networks #5094

Bayesian Networks #5094

Uh oh!

Conversation

tonybruguier commented Mar 17, 2022

Uh oh!

tonybruguier commented Apr 14, 2022

Uh oh!

vtomole commented Apr 14, 2022

Uh oh!

review-notebook-app bot commented Apr 14, 2022

Uh oh!

tonybruguier commented Apr 14, 2022

Uh oh!

MichaelBroughton left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

MichaelBroughton Apr 21, 2022

Choose a reason for hiding this comment

Uh oh!

tonybruguier Apr 22, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

MichaelBroughton Apr 21, 2022

Choose a reason for hiding this comment

Uh oh!

tonybruguier Apr 22, 2022

Choose a reason for hiding this comment

Uh oh!

tonybruguier commented Apr 21, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tonybruguier commented Apr 22, 2022

Uh oh!

tonybruguier commented May 11, 2022

Uh oh!

dabacon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

tonybruguier commented May 16, 2022

Uh oh!

CirqBot commented Jun 8, 2022

Uh oh!

tonybruguier commented Jun 8, 2022

Uh oh!

CirqBot commented Jun 22, 2022

Uh oh!

vtomole commented Jun 22, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

tonybruguier Apr 22, 2022 •

edited

Loading

tonybruguier commented Apr 21, 2022 •

edited

Loading