Feature/implement probability mp.process counts #4943

[PietropaoloFrisoni]

Pull request - issue #4943

Context: This pull request implements the new feature described in issue #4943 (internal assignment).

Description of the Change:
Implemented the method process_counts in the ProbabilityMP class to compute a probability measurement from a counts dictionary.

Benefits:
Representation of store sampling information in the form of a dictionary counts is much more condensed and memory efficient.

Possible Drawbacks:
The new method should reproduce the results of process_samples, implemented in the same class. It has been tested extensively, although I still need to become an expert in quantum computing and circuits.
The function checks that the provided sampling information is a dictionary matching the format returned by CountsMP.

Related GitHub Issues:
#4943

---

Further details:

First of all, congratulations on the library and your fantastic work!

I had the time to explore and play with the code, which is quite amazing. I hope to be considered worthy and help you maintain, expand, and improve it.

I am confident you have improvements and interesting suggestions regarding my pull request. I left some questions for you below as a comment on this PR.

Since this is an internal assignment (for which I was given a deadline), I include some information below that might be relevant for your evaluation (time zone: GMT-5):

• Start time: December 15th, 9:07 AM
• Marked as ready for review on: December 18th, 3:53 AM
• Deadline: December 29th, 9:07 AM

I am incredibly grateful for your time, and I hope to learn so much from all of you!
I wish you all the best.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (68a097c) 99.50% compared to head (d0bfeaa) 99.67%.
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4952      +/-   ##
==========================================
+ Coverage   99.50%   99.67%   +0.16%     
==========================================
  Files         390      392       +2     
  Lines       35585    35288     -297     
==========================================
- Hits        35409    35173     -236     
+ Misses        176      115      -61     

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

[PietropaoloFrisoni]

As I mentioned above, I have a few elementary questions for you:

• I implemented a few checks on the counts parameter to verify that it is a dictionary matching the format returned by class:CountsMP, defined in pennylane.measurements.counts.py. I noticed that there are no similar checks in other methods. Is this a deliberate choice? If that's the case, please tell me if I should remove such checks (or if I should add more of them).
• I understand that the sampling information should be provided so that keys correspond to measurement outcomes in binary strings, like 00010, etc., and no other digit is allowed. Please let me know if that's not the case, and I'll change the code.
• Finally, just a comment. Since I pushed the last changes, the test associated with codecov/project seems stuck, although it was perfectly fine in the previous commit(s). I suspect it is most probably a codecov issue.

If you want to test the code and don't want to write something from scratch, I leave you a very basic snippet I used during some of my tests.

I wish you all the best once again!

@qml.qnode(qml.device("default.qubit"))
def circuit():
    qml.Hadamard(wires=1)
    qml.Hadamard(wires=2)
    qml.CNOT(wires=[0, 1])
    return qml.sample(wires=(0, 1, 2))


wires = (0, 1)
wire_order = qml.wires.Wires((1, 0))

samples = circuit(shots=100)
counts = {}

for i in range(samples.shape[0]):
    binary_string = "".join(map(str, samples[i, :]))
    counts[binary_string] = counts.get(binary_string, 0) + 1

print(f"samples = {samples}")
print(f"counts = {counts}\n")

prob_old = qml.probs(wires=wires).process_samples(samples, wire_order=wire_order)
prob_new = qml.probs(wires=wires).process_counts(counts, wire_order=wire_order)
(prob_old == prob_new).all()

[albi3ro]

Sorry about the delay in getting back to you.

This is looking good 🎉

I left two minor comments, but other than that, this is exactly what I was looking for.

[PietropaoloFrisoni]

Hi @albi3ro and @mudit2812.

Thank you so much for your review and your time!

Sorry about the delay in getting back to you.

No worries, I didn't want to bother you with reminders (especially during Christmas time)

This is looking good 🎉

I left two minor comments, but other than that, this is exactly what I was looking for.

Thank you, Christina!

I pushed a commit with your suggestion and removed the asserts as you both suggested.

If you are not ready to approve the PR, please tell me if there's something else you want me to modify.

Merry Christmas to both of you, as well as to everyone else on the team!

[mudit2812]

Regarding the test failure showing up right now. It's unrelated to your PR @PietropaoloFrisoni and there's no need to worry about it. The test validates a finite-shots workflow where the results are non-deterministic. I will resolve this in a different PR. In the mean time, you can re-run CI without having to make new changes by running git commit --allow-empty -m 'Trigger CI'; git pull; git push

[mudit2812]

These changes look good to me. I'm happy to approve once CI is passing.

[mudit2812]

Congratulations @PietropaoloFrisoni ! Everything is passing, and the changes look good to me. Happy to get this in.

[albi3ro]

Looks great!

I hope you've enjoyed the process and the change to learn a little about pennylane 🎉