Track qpu executions and shots on default qubit

[albi3ro]

Example behavior:

def get_totals(tape):
    dev = qml.device('default.qubit')
    with dev.tracker:
        dev.execute(tape)
    return dev.tracker.totals

# commuting expvals
>>> tape = qml.tape.QuantumScript([], [qml.expval(qml.PauliX(0)), qml.expval(qml.PauliX(1))], shots=10)
>>> get_totals(tape)
{'executions': 1, 'shots': 10, 'batches': 1, 'simulations': 1}

# Hamiltonian with computed grouping indices
>>> H = qml.Hamiltonian([1, 1], [qml.PauliX(0), qml.PauliX(1)], grouping_type="qwc")
>>> tape = qml.tape.QuantumScript([], [qml.expval(H)], shots=10)
>>> get_totals(tape)
{'executions': 1, 'shots': 10, 'batches': 1, 'simulations': 1}

# Hamiltonian without computed grouping indices
>>> H = qml.Hamiltonian([1, 1], [qml.PauliX(0), qml.PauliX(1)])
>>> tape = qml.tape.QuantumScript([], [qml.expval(H)], shots=10)
>>> get_totals(tape)
{'executions': 2, 'shots': 20, 'batches': 1, 'simulations': 1}

# combining expvals with computational basis measurements
>>> tape = qml.tape.QuantumScript([], [qml.expval(qml.PauliX(0)), qml.probs(wires=0)], shots=10)
>>> get_totals(tape)
{'executions': 2, 'shots': 20, 'batches': 1, 'simulations': 1}

# classical shadows
>>> tape = qml.tape.QuantumScript([], [qml.shadow_expval(H0), qml.probs(wires=(0, 1))], shots=10)
>>> get_totals(tape)
{'executions': 11, 'shots': 20, 'batches': 1, 'simulations': 1}

# batch size scaling
>>> tape = qml.tape.QuantumScript([qml.RX((1.2, 2.3), wires=0)], [qml.expval(qml.PauliZ(0))], shots=10)
>>> get_totals(tape)
{'executions': 2, 'shots': 20, 'batches': 1, 'simulations': 1}

[codecov]

Codecov Report

All modified lines are covered by tests ✅

Comparison is base (78f8123) 99.63% compared to head (0dcc80f) 99.63%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #4628   +/-   ##
=======================================
  Coverage   99.63%   99.63%           
=======================================
  Files         376      376           
  Lines       33617    33649   +32     
=======================================
+ Hits        33493    33525   +32     
  Misses        124      124           

Files	Coverage Δ
pennylane/devices/default_qubit.py	100.00% <100.00%> (ø)
pennylane/devices/qubit/sampling.py	100.00% <100.00%> (ø)

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

[albi3ro]

[sc-42389]

[timmysilv]

this looks great, but i'm wondering if we should change the naming to match how things were before. executions kinda already meant qpu_executions on DQL, so maybe we can do:

• executions -> simulator_passes
• qpu_executions -> executions

or something like that. if you make this change, tests/test_vqe.py should fail, and I think that's a good thing (in other words it will need updating to mark this PR as complete).

[trbromley]

Thanks @albi3ro!

this looks great, but i'm wondering if we should change the naming to match how things were before. executions kinda already meant qpu_executions on DQL, so maybe we can do:

* `executions` -> `simulator_passes`

* `qpu_executions` -> `executions`

or something like that. if you make this change, tests/test_vqe.py should fail, and I think that's a good thing (in other words it will need updating to mark this PR as complete).

Agreed - would be great not to introduce a breaking change here. Though this way around there is the issue that simulator_passes doesn't make sense on a real QPU. Maybe in that case we set this number to be zero or None.

[timmysilv]

I think the tracker only shows things that you add to it, right? so on real QPUs, you just don't add simulator_passes.

[albi3ro]

@trbromley @timmysilv

For an analytic case, do we want "executions" to match qpu executions or simulator passes?

[trbromley]

@trbromley @timmysilv

For an analytic case, do we want "executions" to match qpu executions or simulator passes?

@albi3ro in the analytic case, I think it's fine if executions is the number of executions that would be used by default.qubit in finite-shots mode or equivalently on a similar QPU device. We can also store simulator_passes.

I think the above is compatible with what we were doing before, i.e., the old default.qubit would show 2 executions in:

import pennylane as qml

dev = qml.device('default.qubit', wires=1)

@qml.qnode(dev)
def f(x):
    qml.RX(x, wires=0)
    return qml.expval(qml.PauliX(0)), qml.expval(qml.PauliY(0)) 

with qml.Tracker(dev) as tracker:
    f(0.8)

assert tracker.totals['executions'] == 2

I think the tracker only shows things that you add to it, right? so on real QPUs, you just don't add simulator_passes.

@timmysilv yes that works, but can we make sure to clearly document this in the Tracker docs?

[timmysilv]

just a comment on the changelog, otherwise lgtm!

[lillian542]

Looks good! There's a couple things I'm not sure about.