Regarding Number of shots in PEC

[Julie-Maria-Raju]

Hallo,
I was trying to apply pec for a classification tasks and used execute_with_pec with source code having 1 shots. Why is it only 1 shots? I tried it with 1024 shots and got little better results than one shots comparing the accuracy. Have anybody tried pec for classification task? Please let me know.

[natestemen]

Thanks for the question, and cool to hear you're trying to use PEC!

Why is it only 1 shots?

I'll need a little more context here as to what you mean. Are you basing your code off an example that uses a single shot? It sounds like you're saying we have something hard coded to take a single shot for PEC circuits, but I cannot find where, if that is the case.

Have anybody tried pec for classification task?

Not that I'm aware of! What dataset are you working with/testing on? I'm curious to know more about the set up here.

[Julie-Maria-Raju]

Sorry it was my fault, I meant execute_with_noise uses one shot and why?

I am working on OrganAMNIST dataset for class of 11. My expectation values are changing but the classification accuracy doesn’t have an improvement. I expected a better result when I am using long computational time to just validate it with pec with trained weights.

[natestemen]

Ah I see what you mean. So the docstring indicates that the simulation is an exact one. I.e. it's computing $\text{tr}(O \rho)$ without any sampling.

mitiq/mitiq/interface/mitiq_qiskit/qiskit_utils.py

Lines 86 to 92 in be6a76c

	def execute_with_noise(
	circuit: QuantumCircuit,
	obs: npt.NDArray[np.complex64],
	noise_model: NoiseModel,
	) -> float:
	"""Simulates the evolution of the noisy circuit and returns
	the exact expectation value of the observable.

This is how you used to do an exact simulation with the AerSimulator, but I cannot find any documentation on it now. Is it working as expected, or are you seeing strange results?

If you want to use shots to emulate hardware more effectively there is also the execute_with_shots_and_noise function in the same file.

[Julie-Maria-Raju]

Yes , when I tried it with one shots it was giving less validation accuracy like 18 and when i used 1024 it is giving 56.5.
I also have doubt , I am now applying 500 samples for PEC and by increasing it does it improve the mitigation?

[natestemen]

Thanks for raising this. I've asked a question over on the qiskit-aer github page (Qiskit/qiskit-aer#2269). This might be a bug!