Add unit tests for simulate_tree_mcm
#6231
Conversation
|
Hello. You may have forgotten to update the changelog!
|
obliviateandsurrender
changed the title
simulate_tree_mcmsimulate_tree_mcm
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #6231 +/- ##
==========================================
+ Coverage 99.58% 99.71% +0.12%
==========================================
Files 444 443 -1
Lines 42367 42020 -347
==========================================
- Hits 42193 41899 -294
+ Misses 174 121 -53 ☔ View full report in Codecov by Sentry. |
Seconding Mudit here, can we verify the numerical correctness where it makes sense? |
|
@mudit2812 and @astralcai I have added numerical assertions to |
tests/devices/qubit/test_simulate.py
Outdated
| @pytest.mark.unit | ||
| @pytest.mark.parametrize("shots", [None, 5500]) | ||
| @pytest.mark.parametrize("postselect", [None, 0]) | ||
| @pytest.mark.parametrize("reset", [False, True]) | ||
| @pytest.mark.parametrize("measure_f", [qml.counts, qml.expval, qml.probs, qml.sample, qml.var]) | ||
| @pytest.mark.parametrize( | ||
| "meas_obj", [qml.Y(0), [1], [1, 0], "mcm", "composite_mcm", "mcm_list"] | ||
| ) | ||
| def test_simple_dynamic_circuit(self, shots, measure_f, postselect, reset, meas_obj): |
There was a problem hiding this comment.
Instead of marking the individual tests as unit, you can mark the whole test class.
tests/devices/qubit/test_simulate.py
Outdated
| @pytest.mark.unit | ||
| def test_tree_traversal_deep_circuit(self): | ||
| """Test that `simulate_tree_mcm` works with circuits with many mid-circuit measurements""" | ||
|
|
||
| n_circs = 500 | ||
| operations = [] | ||
| for _ in range(n_circs): | ||
| operations.extend( | ||
| [ | ||
| qml.RX(1.234, 0), | ||
| (m0 := qml.measure(0, postselect=1)).measurements[0], | ||
| qml.CNOT([0, 1]), | ||
| qml.ops.op_math.Conditional(m0, qml.RZ(1.786, 1)), | ||
| ] | ||
| ) | ||
|
|
||
| qscript = qml.tape.QuantumScript( | ||
| operations, | ||
| [qml.sample(wires=[0, 1]), qml.counts(wires=[0, 1])], | ||
| shots=20, | ||
| ) | ||
|
|
||
| mcms = find_post_processed_mcms(qscript) | ||
| assert len(mcms) == n_circs | ||
|
|
||
| split_circs = split_circuit_at_mcms(qscript) | ||
| assert len(split_circs) == n_circs + 1 | ||
| for circ in split_circs: | ||
| assert not [o for o in circ.operations if isinstance(o, qml.measurements.MidMeasureMP)] | ||
|
|
||
| res = simulate_tree_mcm(qscript, postselect_mode="fill-shots") | ||
| assert len(res[0]) == 20 | ||
| assert isinstance(res[1], dict) and sum(list(res[1].values())) == 20 |
There was a problem hiding this comment.
I don't really understand the purpose of this test. If we check the correctness of simulate_tree_mcm with at least 3 MCMs, then I don't see the point in checking that the results are correct for 500 MCMs.
There was a problem hiding this comment.
This was meant to test that simulate_tree_mcm can handle sufficiently large numbers of mcms. I brought it here from default_qubit_native_mcm and extended it to test the auxiliary functions the tree-traversal algorithm uses for finding mcms that require post-processing and for splitting the tape on those particular mcms.
| for rs1, rs2 in zip(res1, res2): | ||
| prob_dist1, prob_dist2 = rs1, rs2 | ||
| if measure_f in (qml.sample,): | ||
| n_wires = rs1.shape[1] | ||
| prob_dist1, prob_dist2 = np.zeros(2**n_wires), np.zeros(2**n_wires) | ||
| for prob, rs in zip([prob_dist1, prob_dist2], [rs1, rs2]): | ||
| index, count = np.unique( | ||
| np.packbits(rs, axis=1, bitorder="little").squeeze(), return_counts=True | ||
| ) | ||
| prob[index] = count | ||
|
|
||
| assert qml.math.allclose( | ||
| sp.stats.entropy(prob_dist1 + 1e-12, prob_dist2 + 1e-12), 0.0, atol=5e-2 | ||
| ) |
There was a problem hiding this comment.
Just curious how you arrived at this testing strategy? @maliasadi would love your input here as well.
There was a problem hiding this comment.
The motivation here is to calculate the KL divergence (KLD) between the two probability distributions computed from the samples obtained from the two tapes. It gives a measure that shows how different are distributions.
Context: Adds unit tests for the functionality to simulate mid-circuit measurements natively.
Description of the Change:
Benefits:
Possible Drawbacks:
Related GitHub Issues: [sc-71875]