Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Replace pure python loops with numpy where possible in channels.py. #5839

Merged
merged 3 commits into from
Sep 12, 2022
Merged

Replace pure python loops with numpy where possible in channels.py. #5839

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b3f8fcf
Replace pure python loops with numpy where possible in channels.py.
MichaelBroughton Aug 26, 2022
7cf1643
Adam feedback.
MichaelBroughton Sep 12, 2022
af2b540
Merge branch 'master' into faster_channels
CirqBot Sep 12, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
18 changes: 8 additions & 10 deletions cirq-core/cirq/qis/channels.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,11 +50,9 @@ def kraus_to_choi(kraus_operators: Sequence[np.ndarray]) -> np.ndarray:
Choi matrix of the channel specified by kraus_operators.
"""
d = np.prod(kraus_operators[0].shape, dtype=np.int64)
c = np.zeros((d, d), dtype=np.complex128)
for k in kraus_operators:
v = np.reshape(k, d)
c += np.outer(v, v.conj())
return c
k = np.asarray(kraus_operators)
flat_ops = k.reshape((-1, d))
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think you can achieve the effect of asarray followed by reshape with reshape alone

k = np.reshape(kraus_operators, (-1, d))
return np.einsum(..., k, k.conj())

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As a matter of principle, I think we should prefer errors to be raised as soon as they occur and not propagate far away from the cause since if that happens the error is harder to diagnose and fix.

For this reason, I'd rather we be explicit about the expected array shape and say reshape((len(kraus_operators), d)) rather than reshape((-1, d)).

BTW: This -1 hides a concept that has a name: it is the Choi rank (assuming we were indeed pasted in a Kraus, i.e. minimal representation)! You can use this to make the code read well and (integrating the other comment) say:

choi_rank = len(kraus_operators)
k = np.reshape(kraus_operators, (choi_rank, d))
return np.einsum('bi,bj->ij',, k, k.conj())

This makes it clear that b index in the einsum ranges over the Kraus operators.

return np.einsum('bi,bj->ij', flat_ops, flat_ops.conj())


def choi_to_kraus(choi: np.ndarray, atol: float = 1e-10) -> Sequence[np.ndarray]:
Expand Down Expand Up @@ -105,7 +103,8 @@ def choi_to_kraus(choi: np.ndarray, atol: float = 1e-10) -> Sequence[np.ndarray]

w = np.maximum(w, 0)
u = np.sqrt(w) * v
return [k.reshape(d, d) for k in u.T if np.linalg.norm(k) > atol]
keep = np.linalg.norm(u.T, axis=-1) > atol
return [k.reshape(d, d) for k, keep_i in zip(u.T, keep) if keep_i]


def kraus_to_superoperator(kraus_operators: Sequence[np.ndarray]) -> np.ndarray:
Expand Down Expand Up @@ -140,10 +139,9 @@ def kraus_to_superoperator(kraus_operators: Sequence[np.ndarray]) -> np.ndarray:
Superoperator matrix of the channel specified by kraus_operators.
"""
d_out, d_in = kraus_operators[0].shape
m = np.zeros((d_out * d_out, d_in * d_in), dtype=np.complex128)
for k in kraus_operators:
m += np.kron(k, k.conj())
return m
ops_arr = np.asarray(kraus_operators)
m = np.einsum('bij,bkl->ikjl', ops_arr, ops_arr.conj())
return m.reshape((d_out * d_out, d_in * d_in))


def superoperator_to_kraus(superoperator: np.ndarray, atol: float = 1e-10) -> Sequence[np.ndarray]:
Expand Down