CompilationAveraging

Compilation averaging reduces the effect of choosing one particular physical compilation of an experiment. Perceval divides the requested measurement budget over several compilation seeds, runs those sub-computations, and combines their sample counts into one result.

It can be used when the target platform supports a compilation_seed parameter and different valid compilations may have different physical performance:

>>> import perceval as pcvl
>>>
>>> computer.mitigations = [pcvl.CompilationAveraging(repetitions=3)]

The optional starting_seed makes the sequence of compilation seeds reproducible:

>>> mitigation = pcvl.CompilationAveraging(repetitions=3, starting_seed=100)

Practical considerations

  • The command must accept compilation_seed. Otherwise, this mitigation leaves the computation unchanged.

  • repetitions must be a positive integer.

  • When max_samples or max_shots is set, its value must be at least the number of repetitions. The total requested budget is divided between the repetitions rather than repeated in full for each one.

Note

Avoid setting a too high value for the number of repetitions, as compilation can take some time.

class perceval.runtime.error_mitigation.compilation_averaging.CompilationAveraging(repetitions, starting_seed=None)

Reduce sensitivity to a single physical compilation by averaging the results over several compilations.

The requested samples and shots are divided between repetitions sub-computations, each using a different compilation seed. Their sample counts are combined during post-processing. Commands that do not accept a compilation_seed parameter are left unchanged.

Parameters:

  • repetitions (int) – Number of compilations to average. More repetitions require more compilation work.

  • starting_seed (Optional[int]) – Optional first compilation seed. When omitted, choose one randomly.

extend_computation(computation, imperfections)
Parameters:

  • computation (Computation) – The computation asked by the upper layer

  • imperfections (Imperfections) – The Computer imperfections

Return type:

list[Computation]

Returns:

a list of all computations to execute to apply the mitigation