Optimize the HHL algorithm using amplitude amplification #7141
Conversation
I changed the example to simulate two circuits: the first one is the previous HHL circuit in the example without amplitude amplification, and the second one is the HHL circuit with amplitude amplification. The example demonstrates that amplitude amplification obtains an estimate faster (using fewer total calls to simulate.run).
|
I am happy to provide more explanations about the change if it helps with the review. Just let me know! |
There was a problem hiding this comment.
I think it would be good to provide some tests for this class, which would help increase confidence that this large code change is working correctly.
If possible, it might be good to come to the cirq-cync on Wednesdays to discuss this so that everyone understands the changes. If not possible, we could probably discuss over GH comments.
examples/hhl.py
Outdated
| cirq.measure(ancilla, key='a'), | ||
| ] | ||
| ) | ||
| def __init__(self, A, C, t, register_size, *input_prep_gates): |
There was a problem hiding this comment.
Please add types for all the arguments.
examples/hhl.py
Outdated
| results.append(np.concatenate(results_for_param, axis=0)) | ||
| runs.append(num_runs) | ||
|
|
||
| for label, result, num_runs in zip(('X', 'Y', 'Z'), results, runs): |
There was a problem hiding this comment.
I think it might be better and more testable if we return a result object and then print it out after it returns.
There was a problem hiding this comment.
Done. I moved this to a separate method print_results.
examples/hhl.py
Outdated
|
|
||
| return circuit | ||
|
|
||
| def simulate_without_amplification(self, total_runs): |
There was a problem hiding this comment.
missing type annotations.
examples/hhl.py
Outdated
|
|
||
| return c | ||
|
|
||
| def measure_circuit(self, circuit): |
examples/hhl.py
Outdated
| c.append(cirq.X.on_each(qubits)) | ||
| return c | ||
|
|
||
| def amplitude_amplification(self, num_iterations): |
Thank you! I joined the cirq-dev group (my name is Codrut). Please send me the meeting invite, I'll come. |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #7141 +/- ##
=======================================
Coverage 98.65% 98.65%
=======================================
Files 1106 1106
Lines 95869 95939 +70
=======================================
+ Hits 94581 94651 +70
Misses 1288 1288 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
Thank you for the comments! I added type annotations and created a separate method to print the results. I can add tests next. However, I don't fully understand the initial example, which prevents me from writing some unit tests. How is PhasedXPowGate used to measure X, Y and Z? When I replace the exponent parameters I don't get the matrices for Pauli X, Y or Z. Furthermore, I am not sure how the expected[] array in the example was derived. @verult it would help me a lot if you could give me some hints here. |
This addresses #2216 .
Example output from a run:
To highlight the effect of amplitude amplification, I run the algorithm twice, once without amplification, and once with, targeting the same number of estimates. Amplitude amplification reduces the number of measurements needed for an estimate (from O(1/p) to O(1/sqrt{p})). This shows up as fewer simulation runs.