[Bug] Exclude Snapshot from adjoint backwards pass #2289

dime10 · 2022-03-07T18:28:42Z

The adjoint diff-method on QubitDevices bypasses the normal dev.apply function and instead invokes dev._apply_operation for all operations on the tape. This leads to a bug with the Snapshot operation as implemented on the default.qubit device.
See the following example:

dev = qml.device("default.qubit", wires=1)

@qml.qnode(dev, diff_method="adjoint")
def circuit():
    qml.Hadamard(0)
    qml.Snapshot()
    return qml.expval(qml.PauliZ(0))

circuit()

ValueError: cannot reshape array of size 4 into shape ()

github-actions · 2022-03-07T18:29:02Z

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.md with:

A one-to-two sentence description of the change. You may include a small working example for new features.
A link back to this PR.
Your name (or GitHub username) in the contributors section.

codecov · 2022-03-07T18:52:38Z

Codecov Report

Merging #2289 (62448fd) into v0.22.0-rc0 (d201dd5) will not change coverage.
The diff coverage is 100.00%.

@@             Coverage Diff              @@
##           v0.22.0-rc0    #2289   +/-   ##
============================================
  Coverage        99.31%   99.31%           
============================================
  Files              241      241           
  Lines            19089    19089           
============================================
  Hits             18958    18958           
  Misses             131      131

Impacted Files	Coverage Δ
pennylane/_qubit_device.py	`98.72% <100.00%> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d201dd5...62448fd. Read the comment docs.

antalszava

Looks good, thanks for the quick fix! 🎉

antalszava · 2022-03-07T19:03:21Z

pennylane/_qubit_device.py

@@ -942,7 +942,7 @@ def adjoint_jacobian(self, tape, starting_state=None, use_device_state=False):
                        'the "adjoint" differentiation method'
                    )
            else:
-                if op.name not in ("QubitStateVector", "BasisState"):
+                if op.name not in ("QubitStateVector", "BasisState", "Snapshot"):


Out of scope, but wondering: would this be an issue with Barrier too? 🤔

Hmm good question, I'm not sure about other (hardware?) qubit devices, but for our simulator Barrier is an unsupported operation so it is expanded away.

mlxd

Nice one. Thanks @dime10

dime10 · 2022-03-07T23:24:10Z

@antalszava CodeFactor seems hung up. I ran pylint locally, no new issues.

@josh146

* [Bug] Exclude Snapshot from adjoint backwards pass (#2289) * Exclude Snapshot from adjoint backwards pass * Add snapshots test for diff_methods * Changelog * Trigger rebuild Co-authored-by: antalszava <antalszava@gmail.com> * Work on consistency of `Operator`s (#2287) * some inconsistencies * swap basis * undo duplicated wire in test * changelog * revert snapshot wires change * unused import * merge rc * revert accidental changelog merge Co-authored-by: Josh Izaac <josh146@gmail.com> * Batch run (#2069) * batching ability for non-trainable inputs only following issue #2037 This function creates multiple circuit executions for batched input examples and executes all batch inputs with the same trainable variables. The main difference between the proposed version in the issue and this commit is the input `argnum` this indicates the location of the given input hence gives the ability to work across platforms. * adaptation for batch execution * improvements according to PR rules * minor update according to PR errors * modify according to codefactor-io * reformatted code style * adjust line lenght for linting * update linting * disable linting for too many arguments * add testing for batch input in keras * format test_keras.py * add tests for remaining functions * adapt the defaults * update docstring according to @josh146 's suggestions * remove keras sterilazation * add batch_input to the docstring * docstring update for readability: pennylane/transforms/batch_input.py Co-authored-by: Josh Izaac <josh146@gmail.com> * minor fix in documentation * change assertion error to valueerror * test valueerror * modify the definition of argnum * change argnum -> batch_idx * update changelog-dev.md * apply @josh146 's suggestions * linting * tests * more Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Circuit cutting: Tidy up documentation (#2279) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Apply suggestions from code review Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Add link to communication graph * Reword * Move around * fix * fix Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Circuit cutting: update changelog (#2290) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Update changelog * Link to docs page * Update wording * Apply suggestions from code review Co-authored-by: Josh Izaac <josh146@gmail.com> * Move * Update doc/releases/changelog-dev.md Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Remove * Update Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Minor gradient fixes (#2299) * Produce consisten output shapes In the absence of trainable params, some gradient transforms did not produce an empty tuple yet like the rest of our functions. * Minor formatting changes in param_shift_hessian * Fix param_shift_hessian for all zero diff_methods * Fix missing requires_grad & catch expected warning * Changelog Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> Co-authored-by: David Ittah <dime10@users.noreply.github.com> Co-authored-by: David Wierichs <davidwierichs@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Jack Y. Araz <jackaraz@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca>

@josh146

* [Bug] Exclude Snapshot from adjoint backwards pass (#2289) * Exclude Snapshot from adjoint backwards pass * Add snapshots test for diff_methods * Changelog * Trigger rebuild Co-authored-by: antalszava <antalszava@gmail.com> * Work on consistency of `Operator`s (#2287) * some inconsistencies * swap basis * undo duplicated wire in test * changelog * revert snapshot wires change * unused import * merge rc * revert accidental changelog merge Co-authored-by: Josh Izaac <josh146@gmail.com> * Batch run (#2069) * batching ability for non-trainable inputs only following issue #2037 This function creates multiple circuit executions for batched input examples and executes all batch inputs with the same trainable variables. The main difference between the proposed version in the issue and this commit is the input `argnum` this indicates the location of the given input hence gives the ability to work across platforms. * adaptation for batch execution * improvements according to PR rules * minor update according to PR errors * modify according to codefactor-io * reformatted code style * adjust line lenght for linting * update linting * disable linting for too many arguments * add testing for batch input in keras * format test_keras.py * add tests for remaining functions * adapt the defaults * update docstring according to @josh146 's suggestions * remove keras sterilazation * add batch_input to the docstring * docstring update for readability: pennylane/transforms/batch_input.py Co-authored-by: Josh Izaac <josh146@gmail.com> * minor fix in documentation * change assertion error to valueerror * test valueerror * modify the definition of argnum * change argnum -> batch_idx * update changelog-dev.md * apply @josh146 's suggestions * linting * tests * more Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Circuit cutting: Tidy up documentation (#2279) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Apply suggestions from code review Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Add link to communication graph * Reword * Move around * fix * fix Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Circuit cutting: update changelog (#2290) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Update changelog * Link to docs page * Update wording * Apply suggestions from code review Co-authored-by: Josh Izaac <josh146@gmail.com> * Move * Update doc/releases/changelog-dev.md Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Remove * Update Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Minor gradient fixes (#2299) * Produce consisten output shapes In the absence of trainable params, some gradient transforms did not produce an empty tuple yet like the rest of our functions. * Minor formatting changes in param_shift_hessian * Fix param_shift_hessian for all zero diff_methods * Fix missing requires_grad & catch expected warning * Changelog Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> * Deprecate jacobian tape (#2306) * Deprecate the Jacobian tape * Deprecate tape subclasses * changelog * more test fixes * tests * Apply suggestions from code review Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: antalszava <antalszava@gmail.com> * `qml.generator` doc fixes (#2309) * generator doc fixes * more fixing * Snapshot: remove temporary fixes for lightning device (#2291) * Remove temp fixes for lightning * Include diff_method tests for all devices * Changelog * Update CI to use pennylane-lightning dev Co-authored-by: antalszava <antalszava@gmail.com> * Docs fixes for v0.22.0 release (#2312) * Fix Operator docstring hyperrefs * Fix example for top-level matrix function * Add example to Snapshot op docstring * Fix tape drawing examples in docs * Apply suggestions from code review * Update pennylane/ops/snapshot.py Co-authored-by: Christina Lee <christina@xanadu.ai> * Extend the conditional operations documentation (#2294) * Add qfunc and else to cond's UsageDetails * copy when inverting MV under the hood; add equivalent test case for inversion; add err msg when calling == of MV with unexpected typed obj; more examples * format * test docstr * format * correct examples * format * docstring * have #2300 on rc too * lambda example * intro extend, docstring * changelog PR num * link * note update * updates * Apply suggestions from code review * updates Co-authored-by: Christina Lee <christina@xanadu.ai> * Add `qml.generator(op)` backwards compatibility (#2305) * Add qml.generator(op) backwards compatibility * Apply suggestions from code review Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fix docstring Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fixed docs * use better function refs * pin pennylane-lightning version in CI (#2318) * Amend docstring examples for `compute_matrix` and `compute_eigvals` (#2314) * array() * print() to get the output formatting correct * revert array() * print() Co-authored-by: Maria Schuld <mariaschuld@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Few docstring updates in prep for `v0.22.0` (#2311) * updates * - lower * updates * contract_tensor ref updates * rename test file: batch_input * explicit requires_grad upon param generation * torch.Tensor as type * Update pennylane/transforms/__init__.py Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * use tf in example Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * Support for controlled & adjoint in Snapshot/Barrier (#2315) * controlled, adjoint * Remove print * Match adjoint signature of parent class * Add tests for ctrl/adj support * Update Barrier.adjoint signature * Add ctrl test for Barrier * Update tests/test_debugging.py * Update tests/ops/test_snapshot.py * Update tests/ops/test_snapshot.py * changelog * trigger build Co-authored-by: Antal Szava <antalszava@gmail.com> * ControlledQubitUnitary should raise `DecompositionUndefinedError` (#2320) * DecompositionUndefinedError * changelog * trigger check * `v0.22.0` release notes (#2303) * version * log ref * rename * sections; emojis * format * improvements order * format * addition; collabs; v0.21.0 collab alphabet fix * reorder * collab; deprecation item * more PRs; collab list extended * update * sections * op section break up * correct matrix example * suggestions * suggestions * a few more * fix typo in code * update * no tf import * update Co-authored-by: Josh Izaac <josh146@gmail.com> * don't pull test Lightning (requires v0.22.0 to land) * require v0.22 Lightning or higher * changelog list extend * update * Pin Lightning `>=0.22` (#2324) * pin lightning >=0.22 * Update tests.yml Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: David Ittah <dime10@users.noreply.github.com> Co-authored-by: David Wierichs <davidwierichs@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Jack Y. Araz <jackaraz@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> Co-authored-by: Christina Lee <christina@xanadu.ai> Co-authored-by: Maria Schuld <mariaschuld@gmail.com>

@josh146

* [Bug] Exclude Snapshot from adjoint backwards pass (#2289) * Exclude Snapshot from adjoint backwards pass * Add snapshots test for diff_methods * Changelog * Trigger rebuild Co-authored-by: antalszava <antalszava@gmail.com> * Work on consistency of `Operator`s (#2287) * some inconsistencies * swap basis * undo duplicated wire in test * changelog * revert snapshot wires change * unused import * merge rc * revert accidental changelog merge Co-authored-by: Josh Izaac <josh146@gmail.com> * Batch run (#2069) * batching ability for non-trainable inputs only following issue #2037 This function creates multiple circuit executions for batched input examples and executes all batch inputs with the same trainable variables. The main difference between the proposed version in the issue and this commit is the input `argnum` this indicates the location of the given input hence gives the ability to work across platforms. * adaptation for batch execution * improvements according to PR rules * minor update according to PR errors * modify according to codefactor-io * reformatted code style * adjust line lenght for linting * update linting * disable linting for too many arguments * add testing for batch input in keras * format test_keras.py * add tests for remaining functions * adapt the defaults * update docstring according to @josh146 's suggestions * remove keras sterilazation * add batch_input to the docstring * docstring update for readability: pennylane/transforms/batch_input.py Co-authored-by: Josh Izaac <josh146@gmail.com> * minor fix in documentation * change assertion error to valueerror * test valueerror * modify the definition of argnum * change argnum -> batch_idx * update changelog-dev.md * apply @josh146 's suggestions * linting * tests * more Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Circuit cutting: Tidy up documentation (#2279) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Apply suggestions from code review Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Add link to communication graph * Reword * Move around * fix * fix Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Circuit cutting: update changelog (#2290) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Update changelog * Link to docs page * Update wording * Apply suggestions from code review Co-authored-by: Josh Izaac <josh146@gmail.com> * Move * Update doc/releases/changelog-dev.md Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Remove * Update Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Minor gradient fixes (#2299) * Produce consisten output shapes In the absence of trainable params, some gradient transforms did not produce an empty tuple yet like the rest of our functions. * Minor formatting changes in param_shift_hessian * Fix param_shift_hessian for all zero diff_methods * Fix missing requires_grad & catch expected warning * Changelog Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> * Deprecate jacobian tape (#2306) * Deprecate the Jacobian tape * Deprecate tape subclasses * changelog * more test fixes * tests * Apply suggestions from code review Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: antalszava <antalszava@gmail.com> * `qml.generator` doc fixes (#2309) * generator doc fixes * more fixing * Snapshot: remove temporary fixes for lightning device (#2291) * Remove temp fixes for lightning * Include diff_method tests for all devices * Changelog * Update CI to use pennylane-lightning dev Co-authored-by: antalszava <antalszava@gmail.com> * Docs fixes for v0.22.0 release (#2312) * Fix Operator docstring hyperrefs * Fix example for top-level matrix function * Add example to Snapshot op docstring * Fix tape drawing examples in docs * Apply suggestions from code review * Update pennylane/ops/snapshot.py Co-authored-by: Christina Lee <christina@xanadu.ai> * Extend the conditional operations documentation (#2294) * Add qfunc and else to cond's UsageDetails * copy when inverting MV under the hood; add equivalent test case for inversion; add err msg when calling == of MV with unexpected typed obj; more examples * format * test docstr * format * correct examples * format * docstring * have #2300 on rc too * lambda example * intro extend, docstring * changelog PR num * link * note update * updates * Apply suggestions from code review * updates Co-authored-by: Christina Lee <christina@xanadu.ai> * Add `qml.generator(op)` backwards compatibility (#2305) * Add qml.generator(op) backwards compatibility * Apply suggestions from code review Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fix docstring Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fixed docs * use better function refs * pin pennylane-lightning version in CI (#2318) * Amend docstring examples for `compute_matrix` and `compute_eigvals` (#2314) * array() * print() to get the output formatting correct * revert array() * print() Co-authored-by: Maria Schuld <mariaschuld@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Few docstring updates in prep for `v0.22.0` (#2311) * updates * - lower * updates * contract_tensor ref updates * rename test file: batch_input * explicit requires_grad upon param generation * torch.Tensor as type * Update pennylane/transforms/__init__.py Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * use tf in example Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * Support for controlled & adjoint in Snapshot/Barrier (#2315) * controlled, adjoint * Remove print * Match adjoint signature of parent class * Add tests for ctrl/adj support * Update Barrier.adjoint signature * Add ctrl test for Barrier * Update tests/test_debugging.py * Update tests/ops/test_snapshot.py * Update tests/ops/test_snapshot.py * changelog * trigger build Co-authored-by: Antal Szava <antalszava@gmail.com> * ControlledQubitUnitary should raise `DecompositionUndefinedError` (#2320) * DecompositionUndefinedError * changelog * trigger check * `v0.22.0` release notes (#2303) * version * log ref * rename * sections; emojis * format * improvements order * format * addition; collabs; v0.21.0 collab alphabet fix * reorder * collab; deprecation item * more PRs; collab list extended * update * sections * op section break up * correct matrix example * suggestions * suggestions * a few more * fix typo in code * update * no tf import * update Co-authored-by: Josh Izaac <josh146@gmail.com> * Pin Lightning `>=0.22` (#2324) * pin lightning >=0.22 * Update tests.yml Co-authored-by: Josh Izaac <josh146@gmail.com> * bump the version to 0.22.1 * v0.22.1 release notes * Fix queuing unexpected operators with qml.measure; changelog * notes Co-authored-by: David Ittah <dime10@users.noreply.github.com> Co-authored-by: David Wierichs <davidwierichs@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Jack Y. Araz <jackaraz@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> Co-authored-by: Christina Lee <christina@xanadu.ai> Co-authored-by: Maria Schuld <mariaschuld@gmail.com> Co-authored-by: Guillermo Alonso-Linaje <65235481+KetpuntoG@users.noreply.github.com>

@josh146

…product ops (#2276) * sorted wires to match eig vals when computing expectation value! * Added tests and change log entry * lint * remove redundant lines of code * small typo * [Bug] Exclude Snapshot from adjoint backwards pass (#2289) * Exclude Snapshot from adjoint backwards pass * Add snapshots test for diff_methods * Changelog * Trigger rebuild Co-authored-by: antalszava <antalszava@gmail.com> * Work on consistency of `Operator`s (#2287) * some inconsistencies * swap basis * undo duplicated wire in test * changelog * revert snapshot wires change * unused import * merge rc * revert accidental changelog merge Co-authored-by: Josh Izaac <josh146@gmail.com> * Batch run (#2069) * batching ability for non-trainable inputs only following issue #2037 This function creates multiple circuit executions for batched input examples and executes all batch inputs with the same trainable variables. The main difference between the proposed version in the issue and this commit is the input `argnum` this indicates the location of the given input hence gives the ability to work across platforms. * adaptation for batch execution * improvements according to PR rules * minor update according to PR errors * modify according to codefactor-io * reformatted code style * adjust line lenght for linting * update linting * disable linting for too many arguments * add testing for batch input in keras * format test_keras.py * add tests for remaining functions * adapt the defaults * update docstring according to @josh146 's suggestions * remove keras sterilazation * add batch_input to the docstring * docstring update for readability: pennylane/transforms/batch_input.py Co-authored-by: Josh Izaac <josh146@gmail.com> * minor fix in documentation * change assertion error to valueerror * test valueerror * modify the definition of argnum * change argnum -> batch_idx * update changelog-dev.md * apply @josh146 's suggestions * linting * tests * more Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * removed ordering in * re-push to run tests * push * debugging * Circuit cutting: Tidy up documentation (#2279) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Apply suggestions from code review Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Add link to communication graph * Reword * Move around * fix * fix Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Circuit cutting: update changelog (#2290) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Update changelog * Link to docs page * Update wording * Apply suggestions from code review Co-authored-by: Josh Izaac <josh146@gmail.com> * Move * Update doc/releases/changelog-dev.md Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Remove * Update Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Minor gradient fixes (#2299) * Produce consisten output shapes In the absence of trainable params, some gradient transforms did not produce an empty tuple yet like the rest of our functions. * Minor formatting changes in param_shift_hessian * Fix param_shift_hessian for all zero diff_methods * Fix missing requires_grad & catch expected warning * Changelog Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> * Deprecate jacobian tape (#2306) * Deprecate the Jacobian tape * Deprecate tape subclasses * changelog * more test fixes * tests * Apply suggestions from code review Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: antalszava <antalszava@gmail.com> * `qml.generator` doc fixes (#2309) * generator doc fixes * more fixing * Snapshot: remove temporary fixes for lightning device (#2291) * Remove temp fixes for lightning * Include diff_method tests for all devices * Changelog * Update CI to use pennylane-lightning dev Co-authored-by: antalszava <antalszava@gmail.com> * Docs fixes for v0.22.0 release (#2312) * Fix Operator docstring hyperrefs * Fix example for top-level matrix function * Add example to Snapshot op docstring * Fix tape drawing examples in docs * Apply suggestions from code review * Update pennylane/ops/snapshot.py Co-authored-by: Christina Lee <christina@xanadu.ai> * Extend the conditional operations documentation (#2294) * Add qfunc and else to cond's UsageDetails * copy when inverting MV under the hood; add equivalent test case for inversion; add err msg when calling == of MV with unexpected typed obj; more examples * format * test docstr * format * correct examples * format * docstring * have #2300 on rc too * lambda example * intro extend, docstring * changelog PR num * link * note update * updates * Apply suggestions from code review * updates Co-authored-by: Christina Lee <christina@xanadu.ai> * Add `qml.generator(op)` backwards compatibility (#2305) * Add qml.generator(op) backwards compatibility * Apply suggestions from code review Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fix docstring Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fixed docs * use better function refs * I think I have solved the issue * removed prints * clean up * more cleaning * pin pennylane-lightning version in CI (#2318) * Added permutation on probability vector * Finally fixed issue, just need to add tests and format * general clean up * More clean up * added comment explaining prob vect permutation * added tests for new device method get_ordered_subset() * Update .github/workflows/tests.yml * Update pennylane/_qubit_device.py * Added more tests * Lint * Added changelog entry, fixed up tests * Added PR link to changelog * lint * fixed tests * typo in teests * Fixed get_ordered_subset() to raise a more useful error message * regex match error in tests * Update doc/releases/changelog-dev.md * Update pennylane/_device.py Co-authored-by: antalszava <antalszava@gmail.com> * Update doc/releases/changelog-dev.md * Apply suggestions from code review Co-authored-by: antalszava <antalszava@gmail.com> * Update pennylane/_device.py Co-authored-by: antalszava <antalszava@gmail.com> * Renamed device method for getting ordered subset * update name in qubit device * Wrapped permuting logic into private method * updated tests using pytest.parameterize * moved tests to device test suite * Moved tests to device test suite * added doc string to tests * lint and added check for type of mapped wires * updated device test to use tol as a func instead of a val and run on proper device * typo in tests * address codefactor * more code factor * codefactor * override global variable name * lint * Apply suggestions from code review Co-authored-by: antalszava <antalszava@gmail.com> * format * updated comment in _permute_wires func * updated doc strings in test_measaurements.py Co-authored-by: David Ittah <dime10@users.noreply.github.com> Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: David Wierichs <davidwierichs@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Jack Y. Araz <jackaraz@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> Co-authored-by: Christina Lee <christina@xanadu.ai> Co-authored-by: Maria Schuld <mariaschuld@gmail.com>

@josh146

* Fix output shape of batch transforms Remove squeezing from current batch transforms outputs like gradient_transform and hessian_transform, and instead produce the same output shape that is generated in qml.QNode directly at the batch_transform level. * Do contract with cjac=[1] to remove unit dimension * Fix var_param_shift iterating over 0d array * Fix extra dimension in var_param_shift Due to the mask always being 2d, there is an extra dimension for scalar-valued QNodes, even after contraction with the classical jacobian. Adjust the mask's shape to match that of the arrays holding the result values. * Fix mitigate processing_fn * Fix tests expecting unit dimensions * Fix dimensionality in param_shift_cv * Fix metric tensor tape processing * [Bug] Exclude Snapshot from adjoint backwards pass (#2289) * Exclude Snapshot from adjoint backwards pass * Add snapshots test for diff_methods * Changelog * Trigger rebuild Co-authored-by: antalszava <antalszava@gmail.com> * Work on consistency of `Operator`s (#2287) * some inconsistencies * swap basis * undo duplicated wire in test * changelog * revert snapshot wires change * unused import * merge rc * revert accidental changelog merge Co-authored-by: Josh Izaac <josh146@gmail.com> * Batch run (#2069) * batching ability for non-trainable inputs only following issue #2037 This function creates multiple circuit executions for batched input examples and executes all batch inputs with the same trainable variables. The main difference between the proposed version in the issue and this commit is the input `argnum` this indicates the location of the given input hence gives the ability to work across platforms. * adaptation for batch execution * improvements according to PR rules * minor update according to PR errors * modify according to codefactor-io * reformatted code style * adjust line lenght for linting * update linting * disable linting for too many arguments * add testing for batch input in keras * format test_keras.py * add tests for remaining functions * adapt the defaults * update docstring according to @josh146 's suggestions * remove keras sterilazation * add batch_input to the docstring * docstring update for readability: pennylane/transforms/batch_input.py Co-authored-by: Josh Izaac <josh146@gmail.com> * minor fix in documentation * change assertion error to valueerror * test valueerror * modify the definition of argnum * change argnum -> batch_idx * update changelog-dev.md * apply @josh146 's suggestions * linting * tests * more Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Fix batch execution documented return type * Remove obsolete safe_squeeze * Circuit cutting: Tidy up documentation (#2279) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Apply suggestions from code review Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Add link to communication graph * Reword * Move around * fix * fix Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> * Circuit cutting: update changelog (#2290) * Redo imports * Update docs * Update wording * Fix ID * Add to docs * Add to docs * Fix * Update docstrings * Use nx.MultiDiGraph * Update fragment_graph * Update graph_to_tape * Update remap_tape_wires * Rename to expand_fragment_tape * Update expand_fragment_tape * Update CutStrategy * Update qcut_processing_fn * Remove note * Update cut_circuit * Work on docs * Add to docs * Update pennylane/transforms/qcut.py * Add to changelog * Move device definition * Mention WireCut * Move details * QCut module * Fix image location * Fix init * Update changelog * Link to docs page * Update wording * Apply suggestions from code review Co-authored-by: Josh Izaac <josh146@gmail.com> * Move * Update doc/releases/changelog-dev.md Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Remove * Update Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> * Minor gradient fixes (#2299) * Produce consisten output shapes In the absence of trainable params, some gradient transforms did not produce an empty tuple yet like the rest of our functions. * Minor formatting changes in param_shift_hessian * Fix param_shift_hessian for all zero diff_methods * Fix missing requires_grad & catch expected warning * Changelog Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> * Deprecate jacobian tape (#2306) * Deprecate the Jacobian tape * Deprecate tape subclasses * changelog * more test fixes * tests * Apply suggestions from code review Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: antalszava <antalszava@gmail.com> * `qml.generator` doc fixes (#2309) * generator doc fixes * more fixing * Move tape result squeezing to each transform The previous placement of the squezing inside the batch transform assumes that all uses of the batch transform are similar to that of the gradient transforms, which is not the case. To avoid breaking other type of transforms with this change, the squeezing is now placed inside each gradient transform. Tapes constructed from QNodes now also carry the `_qfunc_output` attribute. * Fix linting errors * Snapshot: remove temporary fixes for lightning device (#2291) * Remove temp fixes for lightning * Include diff_method tests for all devices * Changelog * Update CI to use pennylane-lightning dev Co-authored-by: antalszava <antalszava@gmail.com> * Don't stack the tape result list, only each squeeze each element * Fix `_qfunc_output` missing from expanded tape * Revert "Fix metric tensor tape processing" This reverts commit 03479ac. * Fix issue when stacking scalars The `np.stack` cannot deal with scalar arrays, which can just be skipped in such cases. This situation can occur in the gradient transforms when the output is a scalar array of type object. * Docs fixes for v0.22.0 release (#2312) * Fix Operator docstring hyperrefs * Fix example for top-level matrix function * Add example to Snapshot op docstring * Fix tape drawing examples in docs * Apply suggestions from code review * Update pennylane/ops/snapshot.py Co-authored-by: Christina Lee <christina@xanadu.ai> * Extend the conditional operations documentation (#2294) * Add qfunc and else to cond's UsageDetails * copy when inverting MV under the hood; add equivalent test case for inversion; add err msg when calling == of MV with unexpected typed obj; more examples * format * test docstr * format * correct examples * format * docstring * have #2300 on rc too * lambda example * intro extend, docstring * changelog PR num * link * note update * updates * Apply suggestions from code review * updates Co-authored-by: Christina Lee <christina@xanadu.ai> * Add `qml.generator(op)` backwards compatibility (#2305) * Add qml.generator(op) backwards compatibility * Apply suggestions from code review Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fix docstring Co-authored-by: David Wierichs <davidwierichs@gmail.com> * fixed docs * use better function refs * Remove `safe_squeeze` tests * pin pennylane-lightning version in CI (#2318) * Amend docstring examples for `compute_matrix` and `compute_eigvals` (#2314) * array() * print() to get the output formatting correct * revert array() * print() Co-authored-by: Maria Schuld <mariaschuld@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> * Few docstring updates in prep for `v0.22.0` (#2311) * updates * - lower * updates * contract_tensor ref updates * rename test file: batch_input * explicit requires_grad upon param generation * torch.Tensor as type * Update pennylane/transforms/__init__.py Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * use tf in example Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> * Support for controlled & adjoint in Snapshot/Barrier (#2315) * controlled, adjoint * Remove print * Match adjoint signature of parent class * Add tests for ctrl/adj support * Update Barrier.adjoint signature * Add ctrl test for Barrier * Update tests/test_debugging.py * Update tests/ops/test_snapshot.py * Update tests/ops/test_snapshot.py * changelog * trigger build Co-authored-by: Antal Szava <antalszava@gmail.com> * ControlledQubitUnitary should raise `DecompositionUndefinedError` (#2320) * DecompositionUndefinedError * changelog * trigger check * `v0.22.0` release notes (#2303) * version * log ref * rename * sections; emojis * format * improvements order * format * addition; collabs; v0.21.0 collab alphabet fix * reorder * collab; deprecation item * more PRs; collab list extended * update * sections * op section break up * correct matrix example * suggestions * suggestions * a few more * fix typo in code * update * no tf import * update Co-authored-by: Josh Izaac <josh146@gmail.com> * Pin Lightning `>=0.22` (#2324) * pin lightning >=0.22 * Update tests.yml Co-authored-by: Josh Izaac <josh146@gmail.com> * Review: comment fixes Co-authored-by: Josh Izaac <josh146@gmail.com> * Review: simplify code Co-authored-by: David Wierichs <davidwierichs@gmail.com> * Add tests demonstrating bug resolution * Changelog * Undo difficult to test change for CV device Co-authored-by: antalszava <antalszava@gmail.com> Co-authored-by: David Wierichs <davidwierichs@gmail.com> Co-authored-by: Josh Izaac <josh146@gmail.com> Co-authored-by: Jack Y. Araz <jackaraz@gmail.com> Co-authored-by: Tom Bromley <49409390+trbromley@users.noreply.github.com> Co-authored-by: anthayes92 <34694788+anthayes92@users.noreply.github.com> Co-authored-by: Nathan Killoran <co9olguy@users.noreply.github.com> Co-authored-by: Jay Soni <jbsoni@uwaterloo.ca> Co-authored-by: Christina Lee <christina@xanadu.ai> Co-authored-by: Maria Schuld <mariaschuld@gmail.com>

Exclude Snapshot from adjoint backwards pass

9dab109

dime10 added 2 commits March 7, 2022 13:32

Add snapshots test for diff_methods

7572704

Changelog

dcd7933

dime10 requested review from antalszava and mlxd March 7, 2022 18:46

antalszava approved these changes Mar 7, 2022

View reviewed changes

antalszava and others added 2 commits March 7, 2022 14:19

Merge branch 'v0.22.0-rc0' into fix_snapshot_adjoint

9339643

Trigger rebuild

62448fd

mlxd approved these changes Mar 7, 2022

View reviewed changes

antalszava merged commit 713448d into v0.22.0-rc0 Mar 8, 2022

antalszava deleted the fix_snapshot_adjoint branch March 8, 2022 01:15

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Bug] Exclude Snapshot from adjoint backwards pass #2289

[Bug] Exclude Snapshot from adjoint backwards pass #2289

dime10 commented Mar 7, 2022

github-actions bot commented Mar 7, 2022

codecov bot commented Mar 7, 2022 •

edited

Loading

antalszava left a comment

antalszava Mar 7, 2022

dime10 Mar 7, 2022

mlxd left a comment

dime10 commented Mar 7, 2022

[Bug] Exclude Snapshot from adjoint backwards pass #2289

[Bug] Exclude Snapshot from adjoint backwards pass #2289

Conversation

dime10 commented Mar 7, 2022

github-actions bot commented Mar 7, 2022

codecov bot commented Mar 7, 2022 • edited Loading

Codecov Report

antalszava left a comment

Choose a reason for hiding this comment

antalszava Mar 7, 2022

Choose a reason for hiding this comment

dime10 Mar 7, 2022

Choose a reason for hiding this comment

mlxd left a comment

Choose a reason for hiding this comment

dime10 commented Mar 7, 2022

codecov bot commented Mar 7, 2022 •

edited

Loading