Add a projector observable to CirqDevice

[wongwsvincent]

Context:
Adding a projector observable to CirqDevice.

Description of the Change:
Implementing projector observable with Cirq's own cirq.ProductState.projector method.

Related GitHub PR:
Work coherently with PR #1356

[codecov]

Codecov Report

Merging #62 (d86c2d8) into master (6fe3a57) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master      #62   +/-   ##
=======================================
  Coverage   99.14%   99.14%           
=======================================
  Files           8        8           
  Lines         349      349           
=======================================
  Hits          346      346           
  Misses          3        3           

Impacted Files	Coverage Δ
pennylane_cirq/cirq_device.py	98.90% <ø> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6fe3a57...d86c2d8. Read the comment docs.

[josh146]

Thanks @wongwsvincent! Before I can approve, would you be able to add a small test for the projector observable, to make sure that it works correctly? Since it will be using Cirq directly to compute the expectation, rather than using the logic you added to QubitDevice (I believe).

I think there are two options here:

• Add a small unit test using the device methods directly

• Write an integration test using a QNode (this would require modifying the requirements.txt file so that the latest master of PennyLane is installed instead, e.g.

  git+https://github.com/PennyLaneAI/pennylane.git

Thank you for adding this!

[wongwsvincent]

@josh146, I have only added a test_projector_expectation function in tests/test_expval.py as a start, as I have trouble setting up qsimcirq on my Mac, so I can't run make test locally and I'm hesitating a bit about whether I should continue committing tests I haven't checked myself locally.

P.S. It would be great if there's any suggestion on how to install qsimcirq on Mac OS, but I understand this is not an installation related to PennyLane library.

[josh146]

Thanks @wongwsvincent!

as I have trouble setting up qsimcirq on my Mac, so I can't run make test locally and I'm hesitating a bit about whether I should continue committing tests I haven't checked myself locally.

This is perfectly fine 🙂 If you want to run the test suite locally, you can explicitly ignore the qsim tests, either by only running the test file you have modified,

pytest tests/test_expval.py -x --tb=short

or by using pytest's test selector command line argument:

pytest tests -x --tb=short -k "not (test_qsim or test_qsimh)"

P.S. It would be great if there's any suggestion on how to install qsimcirq on Mac OS, but I understand this is not an installation related to PennyLane library.

Unfortunately I am not sure - I know that qsim is quite tricky to install, even on linux! qsim might provide a dockerfile to use it with Docker, but apart from that, I am unsure. Note that it is perfectly fine to not run the test locally, and instead rely on CI within the PR :)

[wongwsvincent]

Awesome! Thanks @josh146 , something like pytest tests/test_expval.py -x --tb=short is exactly what I need.
I ran that command and it passed the tests successfully. The same tests as the ones I added in pennylane/devices/tests/test_measurements.py are now included in pennylane-cirq/tests/test_expval.py, pennylane-cirq/tests/test_sample.py and pennylane-cirq/tests/test_var.py.
Hopefully all the tests on CI will be fine, and the PR will be ready to merge.

[josh146]

Thanks @wongwsvincent for the extra push with the tests!

The integration tests are still failing, but this is not the fault of your PR, rather, it looks like another bug in the Cirq plugin has been uncovered. Since that is out of scope here, I may override that failing CI check.

[josh146]

I removed 3.6, since PL no longer supports it!

[josh146]

I also updated this to point to PL master, since qml.Projector is not present in the stable v0.15 release

[wongwsvincent]

Thanks @josh146 for fixing some of the test errors!
There are six remaining errors all complaining about the use of numpy.complex64 vs complex128, but they don't look like they are caused by this PR as you have pointed out.
Please let me know if there's anything I am missing or I can help with this PR.

[josh146]

There are six remaining errors all complaining about the use of numpy.complex64 vs complex128, but they don't look like they are caused by this PR as you have pointed out. Please let me know if there's anything I am missing or I can help with this PR.

Yep, they are orthogonal :) I will merge this PR now @wongwsvincent, which allows us to fix these bugs in a separate PR. I have to admit, off the top of my head I'm unsure what the bug is, it will require some investigation.