[WIP] Add mitigation support using mitiq backend #1783
Conversation
|
Hello. You may have forgotten to update the changelog!
|
Codecov Report
@@ Coverage Diff @@
## master #1783 +/- ##
==========================================
- Coverage 98.90% 98.56% -0.34%
==========================================
Files 206 207 +1
Lines 15403 15497 +94
==========================================
+ Hits 15234 15275 +41
- Misses 169 222 +53
Continue to review full report at Codecov.
|
| def new_executor(updated_tape): | ||
| with QuantumTape() as updated_tape_with_measurements: | ||
| for op in updated_tape.operations: | ||
| op.queue() |
There was a problem hiding this comment.
| op.queue() | |
| qml.apply(op) |
this will be safer than calling queue :)
| def new_executor(updated_tape): | ||
| with QuantumTape() as updated_tape_with_measurements: | ||
| for op in updated_tape.operations: | ||
| op.queue() |
There was a problem hiding this comment.
| op.queue() | |
| qml.apply(op) |
this will be safer than calling queue :)
| return new_tape | ||
|
|
||
|
|
||
| def add_noise_to_tape(tape, noisy_op: Operation, noisy_op_args, position: str = "all"): |
There was a problem hiding this comment.
@trbromley you could also
-
Decorate this with
@qml.single_tape_transform, to help remove some of the boiler plate (it allows you to remove thewith QuantumTape() as noisy_tape:context manager) -
Also provide a
@qml.qfunc_transformversion of this function, that allows it to work on qfuncs
add_noise_to_qfunc = qml.qfunc_transform(add_noise_to_tape)
@add_noise_to_qfunc(qml.AmplitudeDamping, 0.2, position="all")
def circuit():
qml.RX(0.9, wires=0)
qml.RY(0.4, wires=1)
qml.CNOT(wires=[0, 1])
qml.RZ(0.5, wires=0)
qml.RX(0.6, wires=1)
return qml.expval(qml.PauliZ(0))Although I don't know if this is something you would want in your workflow or not :)
| def execute(circuit, **kwargs): | ||
| noisy_circuit = add_noise_to_tape(circuit, noisy_op, noisy_op_args, position) | ||
| return original_execute(noisy_circuit, **kwargs) |
There was a problem hiding this comment.
I'm curious if you got this to work via the expand_fn approach?
dev = deepcopy(dev)
original_expand_fn = dev.expand_fn
def expand_fn(self, circuit, **kwargs):
circuit = original_expand_fn(circuit, **kwargs)
return add_noise_to_tape(circuit, noisy_op, noisy_op_args, position)
dev.expand_fn = types.MethodType(expand_fn, dev)
return dev
| >>> tape.execute(dev) | ||
| [0.97578304] | ||
| """ | ||
| dev = deepcopy(dev) |
There was a problem hiding this comment.
I worry that not all devices might be deepcopyable? Might not be an issue though
| def execute(circuit, **kwargs): | ||
| noisy_circuit = add_noise_to_tape(circuit, noisy_op, noisy_op_args, position) | ||
| return original_execute(noisy_circuit, **kwargs) |
There was a problem hiding this comment.
I'm curious if you got this to work via the expand_fn approach?
dev = deepcopy(dev)
original_expand_fn = dev.expand_fn
def expand_fn(self, circuit, **kwargs):
circuit = original_expand_fn(circuit, **kwargs)
return add_noise_to_tape(circuit, noisy_op, noisy_op_args, position)
dev.expand_fn = types.MethodType(expand_fn, dev)
return dev
|
Closing because superseded by #1813 |
Context:
The mitiq library recently merged-in support for input PennyLane tapes. This PR explores how this functionality can be wrapped to support a standard PennyLane workflow.
Description of the Change:
Adds a new
mitigatemodule (+ making visible in documentation and an incomplete set of tests). The module contains:add_noise_to_tape()andadd_noise_to_device()transforms that allow for quick prototyping of noisy circuits without running on hardware (this could live in thetransformsmodule)mitiq_interface()function for wrapping standard functionality from themitiqlibrary (e.g.execute_with_zne), allowing these functions to accept general PL tapes as input circuits and PL devices as executors.mitigate_device()device transform that adds mitigation automatically to the device usingmitiqas a backend.Example UI:
Here are some examples of how a user may interact with the
mitigatemodule.Using mitigate_device
Using mitiq_interface
Benefits:
Mitigation is available within PennyLane, including for gradients.
Possible Drawbacks:
mitigate_device()transform uses the mitiqexecute_with_*function as a black box. We allowexecute_with_*to handle the transformation of the circuit into multiple circuits (e.g. gate folding) and the composition of the execution results into a mitigated value (e.g. extrapolation). This allows us to be agnostic toward the mitigation method.Related PR:
supersedes #1183