Pauli Noise Model Backend

[daniel-mills-cqc]

This adds a sampling backend which adds coherent Pauli errors to each shot of a circuit. A representative use case would be:

from pytket import OpType
from qermit.noise_model import (
    PauliErrorTranspile,
    TranspilerBackend,
    NoiseModel,
    ErrorDistribution,
)
from pytket.pauli import Pauli
import numpy as np
from qiskit import QuantumCircuit
from pytket.extensions.qiskit import qiskit_to_tk

# Bell state qiskit circuit
qiskit_circuit = QuantumCircuit(2)
qiskit_circuit.h(0)
qiskit_circuit.cx(0,1)
qiskit_circuit.measure_all()

# Conver to pytket circuit
pytket_ciruit = qiskit_to_tk(qiskit_circuit)

# Define error rates
error_distribution_dict = {}
error_distribution_dict[(Pauli.X, Pauli.I)] = 0.1
error_distribution = ErrorDistribution(
    distribution=error_distribution_dict,
    rng=np.random.default_rng(seed=0),
)

# Noise model defines which gates these rates apply to
noise_model = NoiseModel({OpType.CX: error_distribution})

# Build backend from noise model
transpiler = PauliErrorTranspile(noise_model=noise_model)
backend = TranspilerBackend(transpiler=transpiler)

# Run experiemnt
n_shots = 100
handle = backend.process_circuit(circuit=pytket_ciruit, n_shots=n_shots)
result = backend.get_result(handle)

result.get_counts()

NoiseModel itself adds additional functionality to perform simulation of logical errors (errors in the form of Pauli strings) in the case of Clifford circuits.

[sjdilkes]

Neat & very readable, nice! I've suggestion some pythonic code edits, typos and some other small suggestions.

One thing to note is that apply_clifford_basis_change_tensor has now been exposed from core pytket, which also considers phase when apply Cliffords to Paulis - this might be worth integrating at some point in the future (doesn't seem like a necessary hold up at this point though!).

[sjdilkes]

noise modelling and simulation*

[sjdilkes]

Can these two checks be replaced with this?

if set(self.distribution.keys()) != set(other.distribution.keys()):
        return False```
        

[sjdilkes]

Can replace the construction of merged_distribution using this I think:

from collections import Counter

return cls(distribution = {
    error: sum(distribution.distribution.get(error, 0) for distribution in distribution_list) / len(distribution_list)
    for error in set(error for dist in distribution_list for error in dist.distribution)
})

[sjdilkes]

If my suggestion doesn't work, I'd type this: Dict[Tuple[Pauli, ...], float]

[sjdilkes]

and move this in to for error in set(sum( ....

[sjdilkes]

Looks like something weird has happened with the formatting in this file? the get_counts method has gen_transpiled_circuit inside its definition?

[daniel-mills-cqc]

It does indeed, and that's intentional. I don't want to expose gen_transpiled_circuit as a class method as the circuit generated is not of use outside of the get counts method. If the user wants to generate a noisy circuit themselves they should use the transpiiler.

[sjdilkes]

I'd reccomend moving gen_transpiled_circuit outside of this function definition, but renaming it to _gen_transpiled_circuit, which is standard practice for signifying a class method is for internal use instead of external use

[sjdilkes]

with the code for producing the counter after the function definitions?

[sjdilkes]

Even if gen_transpiled_circuit and gen_batches are only used in get_counts, it's a bit confusing to have them defined only within that functions scope: I'd recommend making them staticmethods of the class

[daniel-mills-cqc]

I don't want them exposed as methods at all, which is why I have done it this way.

[sjdilkes]

Similar comment to before, I'd make them class methods but rename to _gen_batches

[sjdilkes]

Is this still the case?

[sjdilkes]

I like to add an if name == main at the end of the file and call all the files, & add ->None to each function definition.

[daniel-mills-cqc]

I have done this for the test that do not require a temporary directory to be set up. I'm sure that it's possible to run those tests too but I've not been able to figure out how. Do you have any experience with this?

[daniel-mills-cqc]

Thanks for looking this over. I've made those changes, modulo a few comments.

[sjdilkes]

Two things left to do, one being renaming gen_rng_generator to gen_rng (my name suggestion said generator twice!) & moving concealed methods into underscored class methods.

[sjdilkes]

LGTM