Improve Qobj creation function metadata setting #2388
Conversation
qutip/tests/core/test_gates.py
Outdated
| @@ -142,3 +150,49 @@ def test_gate_normalises_pauli_group(self, gate): | |||
| del pauli_gates[i] | |||
| break | |||
| assert len(pauli_gates) == 0 | |||
|
|
|||
|
|
|||
| @pytest.mark.parametrize("dtype", [qutip.data.Dense, qutip.data.CSR]) | |||
There was a problem hiding this comment.
Should we just check the Dia type here too? I guess the with qutip.CoreOptions(default_dtype="dia") around clifford = ... above was to check the Dia case, but I would prefer not to have to stare at the test to tell that Dia is tested too.
There was a problem hiding this comment.
I choose 2 types so we don't only get the default, but not all type since we are not testing the conversion code, just that it is used. Adding dia would just add more test without testing anything new.
| op.to(dtype) | ||
| for op in map( | ||
| partial(reduce, mul), | ||
| product(_powers(E, 3), _powers(X, 2), _powers(S, 4)), | ||
| ) | ||
| ] | ||
| for gate in gates: | ||
| gate.isherm |
There was a problem hiding this comment.
Set the flags, since there are 24 of them with some being hermitian other not, I call the method compute them. Since they are all 2x2, it does not that a significant time.
| return Qobj( | ||
| [ | ||
| [np.cos(phi / 2), -1j * np.sin(phi / 2)], | ||
| [-1j * np.sin(phi / 2), np.cos(phi / 2)], | ||
| ] | ||
| ], | ||
| isherm=(phi % (2 * np.pi) <= settings.core["atol"]), |
There was a problem hiding this comment.
This is nice but I fear that it may again cause problems with jit. I'm fine just leave it None here. The same for other rotations.
| @@ -72,10 +83,11 @@ def cz_gate(*, dtype="csr"): | |||
| result : :class:`.Qobj` | |||
| Quantum object for operator describing the rotation. | |||
| """ | |||
| return qdiags([1, 1, 1, -1], dims=[[2, 2], [2, 2]], dtype=dtype) | |||
| dtype = dtype or settings.core["default_dtype"] or _data.CSR | |||
| return qdiags([1, 1, 1, -1], dims=_DIMS_2_QB, dtype=dtype) | |||
There was a problem hiding this comment.
How does qdiags handle the isherms property? Will it check if the input is real? If so, maybe for those we could manually set the isherms and isunitary.
There was a problem hiding this comment.
When there is only one diagonal, it sets it automatically.
There is only one case in our code base where this is not the case (and it's not in the tests...).
Improving tests for it and adding manual flag setting could make a good first issue.
Co-authored-by: Simon Cross <hodgestar+github@gmail.com>
Description
Many of the operator functions did not set the
ishermflag.Not having the flag set would cause issue with
jax.jit.ishermflag for almost every functions inoperators.pyandgates.py.qdiagsflags are only set when only one diagonal in used.commutator,squeezing)isunitaryflags for all these also.dtypeparameter to use thedefault_dtypesettings.dtypeparameter.