Surround filter #257
Surround filter #257
Conversation
|
|
||
| point = self.normalize_point(point) | ||
| def default_scoring_fn(region: CircuitRegion) -> float: | ||
| return float(sum(op.num_qudits for op in self[region])) |
There was a problem hiding this comment.
Maybe this isn't super important as it's just a default, but this means single- and multi- qudit operations are very similarly important.
There was a problem hiding this comment.
No, this is an important point. I thought going by the number of qudits in a gate would be enough, but your comment made me realize this needs more. For example, a region with two single-qubit gates is equivalent to one with a two-qubit gate. How do you think we should solve this? Multiply by a factor, e.g., x100?
| region = circuit.surround((1, 3), 4) | ||
| assert region.location == CircuitLocation([2, 3, 4, 5]) | ||
|
|
||
| def test_surround_filter_hard(self) -> None: |
There was a problem hiding this comment.
What are some other use cases for filtering? It seems like maybe a MachineModel/topology or gateset check would be a nice extra test to have.
| break | ||
| # Absorb Single-qudit gates | ||
| index = point[0] | ||
| while index < self.num_cycles - 1: |
There was a problem hiding this comment.
Maybe the parts that do "Absorb Single-qudit gates" and "Check for gates in the middle not in region" should be sub-functions on there own.
There was a problem hiding this comment.
Yeah, using the flag "valid_region" gave me that same thought, but I can't see a way that makes it any simpler. How are you thinking to do this?
New features:
Circuit.surroundnow can take a scoring function and a filterCircuit.nextandCircuit.prevcan now also take aCircuitRegionand return the next or prev points of the entire regionBug Fixes:
exclude=TrueCircuit.check_regionmissing bad regions where a gate would be half-in-half-out