Add noise amplification transformer #6665
Conversation
Co-authored-by: Noureldin <[email protected]>
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #6665 +/- ##
=======================================
Coverage 97.82% 97.82%
=======================================
Files 1066 1068 +2
Lines 91864 91919 +55
=======================================
+ Hits 89862 89917 +55
Misses 2002 2002 ☔ View full report in Codecov by Sentry. |
| self, | ||
| p: float | Mapping[tuple[ops.Qid, ops.Qid], float], | ||
| target_gate: ops.Gate = ops.CZ, | ||
| rng: np.random.Generator | None = None, |
There was a problem hiding this comment.
can you move rng to be an argument of __call__?
| Returns: | ||
| The transformed circuit. | ||
|
|
||
| Raises: |
There was a problem hiding this comment.
can you move the validation of p to the constructor? ... this way you don't need to do it here
| import numpy as np | ||
|
|
||
|
|
||
| def _gate_in_moment(gate: ops.Gate, moment: circuits.Moment) -> bool: |
There was a problem hiding this comment.
nit for performance:
return any(op.gate == gate for op in moment)| transformed_circuit_p1 = na.DepolerizingNoiseTransformer(1.0)(circuit) | ||
| assert len(transformed_circuit_p1) == 20 | ||
| transformed_circuit_p0_03 = na.DepolerizingNoiseTransformer(0.03)(circuit) | ||
| assert 10 <= len(transformed_circuit_p0_03) <= 20 |
There was a problem hiding this comment.
isn't the expected len of the circuit num_moment * (1 + p) ? so here we expect ~$10 * 1.03 = 13$ and 20 correspends to the extreme
| "p must either be a float or a mapping from" # pragma: no cover | ||
| + "sorted qubit pairs to floats" # pragma: no cover | ||
| ) # pragma: no cover | ||
| self.p = p |
There was a problem hiding this comment.
you can resolve p here and then use it in call as self.p_func(pair)
| self.p = p | |
| self.p_func = lambda _: p if isinstance(p, (int, float)) else lambda pair: p.get(pair, 0) |
| p_i = self.p_func(pair_sorted_tuple) | ||
| apply = rng.choice([True, False], p=[p_i, 1 - p_i]) | ||
| if apply: | ||
| choices = [ |
There was a problem hiding this comment.
[optional suggestion] this creates 17 objects for each pair, 16 pauli pairs + the numpy array. consider doing it this way
pauli_a_idx = np.choice(4)
if pauli_a_idx == 0:
pauli_b_idx = np.choice(3) + 1
else:
pauli_b_idx = np.choice(4)
paulit_to_apply = paulis[pauli_a_idx](pair[0]), paulis[pauli_b_idx](pair[1])There was a problem hiding this comment.
Thanks, Nour. I agree that this would be more efficient code, but I'm going to leave it as is for now so that I can move on. (It is already very fast as written.)
Adds a tool for adding depolarizing noise after the specified gate, which can be used on hardware for probabilistic error amplification and zero noise extrapolation.