ConfigurableBackendV2 for easily creating mock backendv2 instances #7501
Conversation
|
Example usage: class FakeExampleV2(ConfigurableFakeBackendV2):
"""A mock backendv2"""
def __init__(self):
qubits = list(range(4))
coupling_map = [[0, 1], [0, 2], [0, 3], [1, 2]]
qubit_coordinates = [[0, 1], [1, 0], [1, 1], [1, 2]]
gate_configuration = {}
gate_configuration[IGate] = [(i,) for i in qubits]
# only can do RXGates on qubits 0 and 4
gate_configuration[RXGate] = [
(i,) for i in list(set(qubits).difference([1, 2]))
]
# can do RY on all qubits
gate_configuration[RYGate] = [(i,) for i in qubits]
# can do CZ on all pairs in coupling map
gate_configuration[CZGate] = [(i, j) for i, j in coupling_map]
# can only measure qubits 2,3
measurable_qubits = [2, 3]
super().__init__(
name="mock_example",
description="a mock backend",
n_qubits=len(qubits),
gate_configuration=gate_configuration,
parameterized_gates={RXGate: "theta", RYGate: "theta"},
measurable_qubits=measurable_qubits,
qubit_coordinates=qubit_coordinates,
) |
|
@evmckinney9 Thanks for your valuable contributions! I am sorry for taking so long to give you a response. @IceKhan13 I took a look through the history of |
Pull Request Test Coverage Report for Build 1773260406
💛 - Coveralls |
There was a problem hiding this comment.
Thanks for starting this, it's good to see BackendV2 used here. I left some questions and comments inline about some changes we should make. Besides those can you add a new features release note about this new class (the details on how to do this are documented here: https://github.com/Qiskit/qiskit-terra/blob/main/CONTRIBUTING.md#release-notes ). Also it would be really great if there were some unit tests that were leveraging this class. I realize that until there is run() support in the class we can't do a full path execution test on aer, but at least validating the construction of ConfigurableFakeBackendV2 objects and that we can use them with transpile() would be very useful in this PR.
| return Options(shots=1024) | ||
|
|
||
| def run(self, run_input, **options): | ||
| raise NotImplementedError |
There was a problem hiding this comment.
I think here we probably should have a comment TODO: Remove after #7391 is implemented and this is based on FakeBackendV2. Since I think this class being able to used as a simulation target is useful.
| if std is None: | ||
| std = 0.01 | ||
|
|
||
| if measurable_qubits is None: | ||
| measurable_qubits = list(range(n_qubits)) | ||
|
|
||
| if not isinstance(qubit_t1, list): | ||
| qubit_t1 = np.random.normal( | ||
| loc=qubit_t1 or 113.0, scale=std, size=n_qubits | ||
| ).tolist() | ||
|
|
||
| if not isinstance(qubit_t2, list): | ||
| qubit_t2 = np.random.normal( | ||
| loc=qubit_t1 or 150.2, scale=std, size=n_qubits | ||
| ).tolist() | ||
|
|
||
| if not isinstance(qubit_frequency, list): | ||
| qubit_frequency = np.random.normal( | ||
| loc=qubit_frequency or 4.8, scale=std, size=n_qubits | ||
| ).tolist() | ||
|
|
||
| if not isinstance(qubit_readout_error, list): | ||
| qubit_readout_error = np.random.normal( | ||
| loc=qubit_readout_error or 0.04, scale=std, size=n_qubits | ||
| ).tolist() | ||
|
|
||
| if dt is None: | ||
| dt = 0.2222222222222222e-9 |
There was a problem hiding this comment.
I wonder if there is a way we can de-duplicate this with ConfigurableBackendV1 as this is identical to the previous implementation. It would be easier to maintain if we need to change something if there was only one place to update.
| self, | ||
| name: str, | ||
| description: str, | ||
| n_qubits: int, |
There was a problem hiding this comment.
We've standardized on using num_qubits everywhere in the qiskit api (the only exception was the models in qiskit.providers.models as those were documented externally as part of the payload specification those are based on). So it would be better to be consistent here and use num_qubits instead.
| online_date=datetime.datetime.utcnow(), | ||
| backend_version="0.0.0", |
There was a problem hiding this comment.
These two attributes aren't strictly needed as they're optional in the BackendV2 class. Is there an advantage to populating them, or can we just not set it?
| self.qubit_readout_error = qubit_readout_error | ||
| self.single_qubit_gates = single_qubit_gates | ||
| self.std = std | ||
| # self.n_qubits = n_qubits |
There was a problem hiding this comment.
| # self.n_qubits = n_qubits |
| } | ||
| if gate in parameterized_gates.keys(): | ||
| self._target.add_instruction( | ||
| gate(Parameter(parameterized_gates[gate])), temp_gate_props |
There was a problem hiding this comment.
What about gates with more than 1 parameter. For example, if I want to use a U which has 3 parameters I don't think this would work because I can't specify the 3 different parameter names.
There was a problem hiding this comment.
In instantiation, pass paramerized gates dictionary with values as a list
parameterized_gates={U3Gate: ["theta", "phi", "lam"], RYGate: ["theta"]},| gate(Parameter(parameterized_gates[gate])), temp_gate_props | |
| gate(*[Parameter(p) for p in parameterized_gates[gate]]), | |
| temp_gate_props, | |
| # TODO: dynamic instruction properties | ||
| for gate, qubit_tuple_list in self.gate_configuration.items(): | ||
| temp_gate_props = { | ||
| qubit_tuple: InstructionProperties(duration=0.0, error=0) |
There was a problem hiding this comment.
I think it would be better to use:
| qubit_tuple: InstructionProperties(duration=0.0, error=0) | |
| qubit_tuple: None |
if you're not going to set a duration or error rate. Looking at what the v1 version of this class did they set error to 0.01 and 4*dt for the duration. We could just use that here to start.
| qubit_t2: Optional[Union[float, List[float]]] = None, | ||
| qubit_frequency: Optional[Union[float, List[float]]] = None, | ||
| qubit_readout_error: Optional[Union[float, List[float]]] = None, | ||
| single_qubit_gates: Optional[List[str]] = None, |
There was a problem hiding this comment.
How is this parameter used? It doesn't seem to match up with gate_configuration or parameterized_gates. For example, you could specify completely different gates here and there would be nothing that errors. I'm just wondering if we can remove this argument?
There was a problem hiding this comment.
Looks like this was accidentally left in from an old version - can be removed.
| gate_configuration: Dict[Gate, List[Tuple[int]]], | ||
| measurable_qubits: List[int] = None, | ||
| parameterized_gates: Optional[Dict[Gate, str]] = None, | ||
| qubit_coordinates: Optional[List[List[int]]] = None, |
There was a problem hiding this comment.
How is this parameter used? I don't see it on V1 and am not familiar with a similar parameter on other backends. Is this just specifying a layout for visualization (which is how I interpret the docstring below)? If so where does this get used because I've not seen this parameter anywhere before.
There was a problem hiding this comment.
I was using qubit_coodinates to specify visualization layouts, see #7494
|
@evmckinney9 I have been working on PR #7643 that are closely related to this one. Would you like to update your PR to leverage on the changes in #7643 after it's merged (hopefully in a week). We can discuss together on the changes. |
HuangJunye
added
the
mod: fake_provider
HuangJunye
added
the
Community PR
|
hi @evmckinney9 are you still working on this? |
|
I have not maintained it since writing this PR |
|
Ok thanks for letting us know @evmckinney9, I'll close the PR but if you want to return at a later date just let us know and we can reopen it 😄 |
Summary
Addresses #7499.
Details and comments
This initial commit contains some remaining TODOs.
I used FakeMumbaiV2 and ConfigurableBackend(V1) to create this. I tried to make this simple so you don't have to pass a ton to this factory. The main parameter is in a dictionary that contains the gates and the qubits they can be applied on. The rest is just dictionary comprehension to build the InstructionProperties and QubitProperties.