Add in-place variants for infer_pauli_measurements, to_bloch,...

[thierry-martinez]

Method Pattern.infer_pauli_measurements now operates in place by default, with an optional copy parameter. If copy=True, a modified copy is returned instead.

Added the following methods to Pattern. By default, they modify the pattern in place, with an optional copy parameter. If copy=True, a modified copy is returned instead.

• apply, an in-place variant of map,
• blochify, an in-place variant of to_bloch,
• substitute, an in-place variant of subs,
• replace_parameters, an in-place variant of xreplace.

This is a step toward the cleaner simulation API proposed in #493. It also makes the examples in the Graphix paper cleaner.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.26%. Comparing base (0412960) to head (2cf8a94).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #568      +/-   ##
==========================================
+ Coverage   89.11%   89.26%   +0.15%     
==========================================
  Files          49       49              
  Lines        7716     7791      +75     
==========================================
+ Hits         6876     6955      +79     
+ Misses        840      836       -4     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[matulni]

Thanks for the change, new functions are useful!

I'm not convinced by the naming of the parametric angle functions. The following are not very intuitive to me:

• subs and substitute have the same relationship as xreplace and replace_parameters.
• substitute and replace_parameters act in-place.
• subs replaces one parameter, and xreplace replaces multiple parameters in parallel. (I'm aware that this is a sympy convention and that it's independent of this PR, but I always found it difficult to remember. In addition, with the new methods we drift away from sympy's convention).

Here are two alternatives:

A)
substitute -> subs_in_place
replace_parameters -> xreplace_in_place

It solves 1) and 2), but it's more verbose and not true when calling subs_in_place(..., copy=True).

B)
subs -> substituted

substitute remains

xreplace -> substituted_multi

replace_parameters -> substitute_multi

I think this solves all three points but we don't follow sympy convention at all, and it requires a larger refactoring (we would need to update analogous methods in OpenGraph, etc.). If you think the -ed suffix may be easier to miss when reading the code, we could use with_substitution instead.

What do you think ?

[pranav97nair]

Thanks for this change. The new functions look good to me.

[thierry-martinez]

Thank you, @matulni and @pranav97nair, for your reviews.

I consolidate the API changes:

• All classes that have subs and xreplace methods (Pattern, Circuit, OpenGraph, Measurement, PauliFlow, XZCorrections, DensityMatrix) now derive from the Parameterizable class, which exposes the method with_parameter and with_parameters, with subs and xreplace as their respective aliases.
• Pattern and Circuit now derive from the InplaceParameterizable subclass, which additionally provides the following methods:
  • replace_parameter, an in-place variant of with_parameter (or subs),
  • replace_parameters, an in-place variant of with_parameters (or xreplace).

Moreover, I added a class AbstractStatevector to allow Statevector defined in graphix-symbolic to be pretty-printed like the built-in Statevector.

[matulni]

Thanks a lot for implementing this changes, it's a lot of work.

I'm happy with all the changes regarding the parametric substitution.

However, I believe that the new shared methods in the AbstractStatevector class won't work with the symbolic backend (I give a specific example below). I suspect that the code is not generic enough to work on the GPU backend either.

Making this methods work requires some additional work on the symbolic backend, which I don't think is worth at all. Then, as it is now, I don't think we need the AbstractStatevector class. I agree that in the future it could be practical if we implement others backend more useful than the symbolic one, but can see then.

Something I missed: is the AbstractStatevector class necessary to make the new work on the parametric substitution work ? Couldn't we leave it for a future PR ?

[pranav97nair]

Thanks for these changes, Thierry. The new abstract classes Parameterizable and InplaceParametrizable are nice additions. I've just found a couple minor typos.