Fix PrepSelPrep differentiability with parameter-shift
#6423
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hello. You may have forgotten to update the changelog!
|
PrepSelPrep differentiability with parameter-shiftPrepSelPrep differentiability with parameter-shift
PrepSelPrep differentiability with parameter-shiftPrepSelPrep differentiability with parameter-shift
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #6423 +/- ##
==========================================
- Coverage 99.70% 99.38% -0.32%
==========================================
Files 447 447
Lines 42428 42429 +1
==========================================
- Hits 42303 42169 -134
- Misses 125 260 +135 ☔ View full report in Codecov by Sentry. |
astralcai
reviewed
Oct 22, 2024
mudit2812
pushed a commit
that referenced
this pull request
Nov 11, 2024
**Context:**
Prior to this fix, differentiating `PrepSelPrep` with
`diff_method="parameter-shift"` resulted in zeros (which was
inconsistent with `backprop`),
```python
@qml.qnode(qml.device("default.qubit"), diff-method="parameter-shift")
def circuit(params):
H = qml.Hamiltonian(params, [qml.Z(2), qml.X(1) @ qml.X(2)])
qml.PrepSelPrep(H, control=[0])
return qml.expval(qml.PauliZ(0))
>>> x = qml.numpy.array([0.25, 0.75])
>>> qml.jacobian(circuit)(x)
array([0., 0.])
```
**Description of the Change:**
One line change to add `PrepSelPrep.grad_method=None` 😺. This results
in,
```python
@qml.qnode(qml.device("default.qubit"), diff-method="parameter-shift")
def circuit(params):
H = qml.Hamiltonian(params, [qml.Z(2), qml.X(1) @ qml.X(2)])
qml.PrepSelPrep(H, control=[0])
return qml.expval(qml.PauliZ(0))
>>> x = qml.numpy.array([0.25, 0.75])
>>> qml.jacobian(circuit)(x)
array([-1.5, 0.5])
```
_Reason this works:_ It seems that parameter-shift tries to update the
parameters and fails to do so. Setting `grad_method=None` forces it to
decompose the operator before differentiating it's data. This does
suggest some areas of improvement for `Operator.grad_method`.
The `xfails` were also removed from the relevant tests. Also, certain
test cases would result in `NaN`s upon differentiation with *any*
method, so I skipped those cases as they are a *known* limitation of
`MottonenStatePreparation`.
**Benefits:** Gradient results are now consistent with `backprop`.
**Possible Drawbacks:** None.
**Related GitHub Issues:** Fixes #6331
[sc-74920]
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Context:
Prior to this fix, differentiating
PrepSelPrepwithdiff_method="parameter-shift"resulted in zeros (which was inconsistent withbackprop),Description of the Change:
One line change to add
PrepSelPrep.grad_method=None😺. This results in,Reason this works: It seems that parameter-shift tries to update the parameters and fails to do so. Setting
grad_method=Noneforces it to decompose the operator before differentiating it's data. This does suggest some areas of improvement forOperator.grad_method.The
xfailswere also removed from the relevant tests. Also, certain test cases would result inNaNs upon differentiation with any method, so I skipped those cases as they are a known limitation ofMottonenStatePreparation.Benefits: Gradient results are now consistent with
backprop.Possible Drawbacks: None.
Related GitHub Issues: Fixes #6331
[sc-74920]