Serialization of measurements

[BrunoLiegiBastonLiegi]

This addresses the bug #1426 by implementing a proper serialization of measurement gates and their result.
Furthermore, the possibility to specify the basis of measurement gates with strings (for instance "Z" or ["X", "Z", "Y"]) has been added.
I also made some minor modifications to MeasurementResult, and I would also like to remove the circuit attribute from it. At the moment it is only used to generate the samples when they are not available. However, as I see it, the MeasurementResult should always be associated with the samples.

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 ✅

Project coverage is 99.94%. Comparing base (21732ad) to head (7864590).
Report is 123 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1431   +/-   ##
=======================================
  Coverage   99.94%   99.94%           
=======================================
  Files          78       78           
  Lines       11317    11324    +7     
=======================================
+ Hits        11311    11318    +7     
  Misses          6        6           

Flag	Coverage Δ
unittests	99.94% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

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

[alecandido]

Some suggestions, but they are not incredibly relevant, as the whole measurement storage is a bit convoluted (with many references everywhere), and a greater action should be taken to simplify it (not now, possibly as part of qibo-core).

The relevant part instead it's that a test is missing. It would be nice to have a test failing before this PR, and now working.

[alecandido]

Furthermore, the possibility to specify the basis of measurement gates with strings (for instance "Z" or ["X", "Z", "Y"]) has been added.

Try to keep features' implementation as separate as possible (and separate from bug fixes).

(Ok, this is really a tiny one, but it was worth to mention the general rule)

[BrunoLiegiBastonLiegi]

Yeah, let's say that rather than a new feature is a side effect of the fix in this case, but I fundamentally agree.

[BrunoLiegiBastonLiegi]

Ok, I should have addressed everything now, except the MeasurementResult.circuit attribute that is still there, but I would get rid of it if everyone agrees.

[alecandido]

The test is effective and passing.

Thanks @BrunoLiegiBastonLiegi!

[marcorossi5]

Hello guys, I've installed qibo from this source branch and checked if the qibo-webappdaemon works smoothly.
Unfortunately we are having the following error raised by the following snippet:

with open(circuit_path, "r") as f:
      raw = json.load(f)
c = qibo.Circuit.from_dict(raw)
r = c(nshots=settings.get("nshots"))
print(f"Results: {r.samples()}")

Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/nfs/users/tiiq/webappdaemon/src/daemon/slurm_circuit_computation.py", line 62, in <module>
    main(args.job_folder)
  File "/nfs/users/tiiq/webappdaemon/src/daemon/slurm_circuit_computation.py", line 55, in main
    print(f"Results: {r.samples()}")
  File "/nfs/users/tiiq/qibo/src/qibo/result.py", line 324, in samples
    qubits = self.measurement_gate.target_qubits
AttributeError: 'NoneType' object has no attribute 'target_qubits'

[scarrazza]

Do you have a simpler example using only qibo primitives?

[marcorossi5]

I did a little bit of investigation. It seems that the circuit serialization done with circuit.raw looks different when launched on a CPU only queue (like sim) or on a QPU one (qw11q).

The script is the following:

import json
import qibo

c = qibo.Circuit(5)
c.add(qibo.gates.M(0))

with open("circuit.json", "w") as f:
    json.dump(c.raw, f)

print(c.raw)

Here you have the results:

Output on sim

{'queue': [{'name': 'measure', 'init_args': (0,), 'init_kwargs': {'register_name': None, 'collapse': False, 'basis': ['Z'], 'p0': None, 'p1': None}, '_target_qubits': (0,), '_control_qubits': (), '_class': 'M', 'measurement_result': {'samples': None}}], 'nqubits': 5, 'density_matrix': False, 'qibo_version': '0.2.12'}

Output on qw11q:

{'queue': [{'name': 'measure', 'init_args': [0], 'init_kwargs': {}, '_target_qubits': [0], '_control_qubits': [], '_class': 'M'}], 'nqubits': 5, 'density_matrix': False, 'qibo_version': '0.2.11'}

It seems that there's a mismatch when loading a circuit on the QPU that was created on CPU and viceversa.

The code that raises the following errors is:

import json
import qibo
with open("circuit.json") as f:
    raw = json.load(f)

c = qibo.Circuit.from_dict(raw)

If I create the circuit on CPU => load it on QPU:

Traceback (most recent call last):
  File "/nfs/users/tiiq/from_raw.py", line 6, in <module>
    c = qibo.Circuit.from_dict(raw)
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/models/circuit.py", line 1137, in from_dict
    circ.add(Gate.from_dict(gate))
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/gates/abstract.py", line 109, in from_dict
    gate = cls(*raw["init_args"], **raw["init_kwargs"])
  File "/nfs/tools/qibo/lib/python3.10/site-packages/qibo/gates/measurements.py", line 101, in __init__
    gate = basis_cls(qubit).basis_rotation()
TypeError: 'str' object is not callable
srun: error: fahid: task 0: Exited with exit code 1

If I create the circuit on QPU => load it on CPU:

Traceback (most recent call last):
  File "/nfs/users/tiiq/from_raw.py", line 6, in <module>
    c = qibo.Circuit.from_dict(raw)
  File "/nfs/users/tiiq/qibo/src/qibo/models/circuit.py", line 1137, in from_dict
    circ.add(Gate.from_dict(gate))
  File "/nfs/users/tiiq/qibo/src/qibo/gates/abstract.py", line 121, in from_dict
    if raw["_class"] == "M" and raw["measurement_result"]["samples"] is not None:
KeyError: 'measurement_result'

[BrunoLiegiBastonLiegi]

The output on sim should be the correct one, which was updated in this PR. The other one seems to be the old serialization of the M gate (the one in main).

Similarly below, the CPU -> QPU loading seems to fail because the M object is not able to receive strings as basis arguments, which is an additional feature introduced here. Viceversa, the QPU -> CPU loading fails because the object serialized on the QPU doesn't include attributes that are required to load a measurement now, i.e. the serialization of the MeasurementResult.

I would say that the CPU and QPU are using two different versions of qibo:

• CPU -> this branch
• QPU -> main

[marcorossi5]

Let me check again the environments. If the branch mismatch is the cause, then we are fine.

[scarrazza]

I have just tried your code and it works, no crashes.

[marcorossi5]

Yes, I agree. I was loading the wrong environment on the QPU, my bad.
I guess the test is passed then