Add eigvals support to ExpectationMP, CountsMP, VarianceMP, SampleMP

[astralcai]

Context:
MeasurementProcess supports specifying eigvals instead of obs, which should include sample based measurements such as ExpectationMP, VarianceMP. However, process_samples of these measurements does not work if the measurement was created using eigvals

Description of the Change:
Adds eigvals support to variance sample measurements

Benefits:
Bugfix

Possible Drawbacks:
This fix only makes sure that the process_samples method would work. Specifically for expval, as it is also a state measurement, currently taking a expval in a qnode with only eigvals specified does not work.

Related GitHub Issues:
Fixes #5432
[sc-59519]

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

❗ No coverage uploaded for pull request base (master@f10e98f). Click here to learn what that means.

Additional details and impacted files

@@            Coverage Diff            @@
##             master    #5463   +/-   ##
=========================================
  Coverage          ?   99.66%           
=========================================
  Files             ?      402           
  Lines             ?    37347           
  Branches          ?        0           
=========================================
  Hits              ?    37221           
  Misses            ?      126           
  Partials          ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[lillian542]

Looking good! Left a bit of feedback regarding the wires check, otherwise just a couple very minor comments from me 🚀

[PietropaoloFrisoni]

Fantastic, thanks @astralcai. I notice that some statements in strings have periods, and others do not (and I am not a massive fan of capital words), but it seems to me you just adapted such a style to the existing context.

[albi3ro]

I'm a little hesitant to make specifying eigenvalues user-facing via the constructor functions.

While we do have internal logic for having eigvals+wires as the description for a measurement (see MeasurementProcess.expand), it isn't particularly well supported right now (see PR #5445 for some other places things are going wrong). The bugfix is part of the first push to improving our support for this specification, but we need to more testing and support improvement.

I would recommend breaking this PR into two, one fixing the internal bug, and another making this representation user-facing. In that way, the bug fix won't be blocked by product concerns.

[astralcai]

I'm a little hesitant to make specifying eigenvalues user-facing via the constructor functions.

While we do have internal logic for having eigvals+wires as the description for a measurement (see MeasurementProcess.expand), it isn't particularly well supported right now (see PR #5445 for some other places things are going wrong). The bugfix is part of the first push to improving our support for this specification, but we need to more testing and support improvement.

I would recommend breaking this PR into two, one fixing the internal bug, and another making this representation user-facing. In that way, the bug fix won't be blocked by product concerns.

ExpectationMP.process_samples calls qml.sample to create a SampleMP, and uses SampleMP.process_samples to preprocess the samples. Then it calculates the expectation value by taking the mean of the processed sample. The internal bug was caused by qml.sample not being able to handle eigvals. In order to fix the internal bug, eigvals must be added as an argument to qml.sample, but it can be left out of the other measurements for now. What do you think?

[albi3ro]

I'm a little hesitant to make specifying eigenvalues user-facing via the constructor functions.
While we do have internal logic for having eigvals+wires as the description for a measurement (see MeasurementProcess.expand), it isn't particularly well supported right now (see PR #5445 for some other places things are going wrong). The bugfix is part of the first push to improving our support for this specification, but we need to more testing and support improvement.
I would recommend breaking this PR into two, one fixing the internal bug, and another making this representation user-facing. In that way, the bug fix won't be blocked by product concerns.

ExpectationMP.process_samples calls qml.sample to create a SampleMP, and uses SampleMP.process_samples to preprocess the samples. Then it calculates the expectation value by taking the mean of the processed sample. The internal bug was caused by qml.sample not being able to handle eigvals. In order to fix the internal bug, eigvals must be added as an argument to qml.sample, but it can be left out of the other measurements for now. What do you think?

Could we just directly create a SampleMP?

[trbromley]

+1 to @albi3ro's comment, do we need to add another argument to qml.sample(), etc., to fix this bug?

[astralcai]

qml.sample contains logic that handles the creation of a SampleMP. I can create a helper method to contain that logic. I will make the update.

[albi3ro]

qml.sample contains logic that handles the creation of a SampleMP. I can create a helper method to contain that logic. I will make the update.

I'd be fine with moving more validation and preprocessing into SampleMP.__init__.

[albi3ro]

Thanks for adding this! Looks great :)