Circuit Execution Outputs

[BrunoLiegiBastonLiegi]

As requested in #1020, three objects are implemented that substitute the old CircuitResult object.
Depending on the different conditions, now the circuit execution returns:

1. no noise, no measurements --> QuantumState
2. no noise, measurements --> CircuitResult
3. noise, no measurements --> raise error if density_matrix=False else QuantumState
4. noise, measurements --> CircuitResult if density_matrix=True else MeasurementOutcomes

where the new CircuitResult(QuantumState, MeasurementOutcomes) object contains both the QuantumState and the MeasurementOutcomes.
nshots is set by default to 1000 now.
The probabilities and the state can be recovered as QuantumState.probabilities() and QuantumState.state(), the frequencies and the samples with MeasurementOutcomes.frequencies() and MeasurementOutcomes.samples(). CircuitResult gives access to all of them in the same way.

Currently the only case that has been left out is measurements with collapse=True.
One possibility to handle them could be:

• only collapsing measurements present
  --> QuantumState if density_matrix=True
  --> raise Error if density_matrix=False
• both collapsing and standard measurements present
  --> CircuitResult if density_matrix=True
  --> MeasurementOutcomes if density_matrix=False

Furthermore, handling the Circuit._final_state attribute now is a bit problematic, as depending on the conditions different different things are obtained after execution.

Checklist:

• Reviewers confirm new code works as expected.
• Tests are passing.
• Coverage does not decrease.
• Documentation is updated.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files	Coverage Δ
src/qibo/__init__.py	100.00% <ø> (ø)
src/qibo/backends/abstract.py	100.00% <100.00%> (ø)
src/qibo/backends/numpy.py	100.00% <100.00%> (ø)
src/qibo/backends/tensorflow.py	100.00% <100.00%> (ø)
src/qibo/gates/measurements.py	100.00% <100.00%> (ø)
src/qibo/measurements.py	100.00% <100.00%> (ø)
src/qibo/models/circuit.py	100.00% <100.00%> (ø)
src/qibo/models/error_mitigation.py	100.00% <ø> (ø)
src/qibo/models/qgan.py	100.00% <100.00%> (ø)
src/qibo/models/tsp.py	100.00% <ø> (ø)
... and 3 more

📢 Thoughts on this report? Let us know!.

[BrunoLiegiBastonLiegi]

Okay this should be ready for review now. I am also working on adding a table to the docs to summarize the circuit execution outcomes depending on the settings.

[stavros11]

Thank you for working on this @BrunoLiegiBastonLiegi. The new objects look good and the results are much cleaner now. I wrote some comments below, note that some of them are hidden because they are many. I still need to have a look in the documentation updates and a more detailed look in the test updates.

Other than that, it may be worth updating this with main and fixing the conflicts every now and then because conflicts will block the merge even if we approve. In principle you can fix once when we are ready to merge but is usually better to maintain this up-to-date with main.

[scarrazza]

@BrunoLiegiBastonLiegi thanks for this. Could you please also consider the possibility to store and load the numpy object that are required in order to reallocate the results classes? This is necessary for the future remote connection mechanism based on http requests.

[scarrazza]

Could you please include the qibo version in the results objects and remove the backend from the dump and use numpy as default when loading?

[stavros11]

I had a quick look at the docs as I have not done this before and it looks good to me. Feel free to merge this and open an issue about the probabilities or anything else we could potentially address in the future.

I will test the qibolab part later and if there is any update required here we can fix in a new PR.