Use qml.math.cast after qml.math.convert_like in qml.gradients to copy dtype
#2120
Conversation
|
Hello. You may have forgotten to update the changelog!
|
antalszava
changed the title
qml.math.cast_like instead of qml.math.convert_like in qml.gradientsqml.math.cast after qml.math.convert_like in qml.gradients to copy dtype
Codecov Report
@@ Coverage Diff @@
## master #2120 +/- ##
=======================================
Coverage 99.19% 99.19%
=======================================
Files 228 228
Lines 17476 17485 +9
=======================================
+ Hits 17335 17344 +9
Misses 141 141
Continue to review full report at Codecov.
|
| @@ -46,6 +46,7 @@ def compute_vjp(dy, jac, num=None): | |||
|
|
|||
| if not isinstance(dy_row, np.ndarray): | |||
| jac = math.convert_like(jac, dy_row) | |||
| jac = math.cast(jac, dy_row.dtype) | |||
There was a problem hiding this comment.
very minor @antalszava, but I'm curious why cast_like wasn't used here, I thought it would be safer that using dy_row.dtype?
There was a problem hiding this comment.
qml.math.cast_like assumes that the two tensors are already of the same type if I'm not mistaken. I ran into issues using it with Torch, see this commit: 8d84766
|
[sc-13632] |
Context:
qml.math.cast_likeattempts to copy the dtype of a tensor whereasqml.math.convert_likeonly uses the type of the tensors during conversionThe
qml.gradientsmodule usesqml.math.convert_like. There may, however, be cases where the dtype would need to be copied, otherwise unexpected behaviour arises.Description of the Change:
Benefits:
Possible Drawbacks:
Related GitHub Issues:
#2105