Bugfix reparameterize #480

juangmendoza19 · 2024-08-23T18:12:24Z

Addressing issue #425.

convert() function has been changed for POVM elements and state preparation. They now support proper conversion from full TP to GLND parameterization. The optimization is now done over non-gauge directions, which are found through Jacobians.

This function still needs to be changed to appropriately handle other reparameterizations.

Lastly, this is not guaranteed to work for POVMs, just because the number of degrees of freedom of a POVM can be greater than the number of error generators that span TP maps for a given number of qubits. By pigeon hole principle, we may not be able to find a description for a POVM consisting of Ideal_POVM + error generators. There is an assertion to try to catch this.

…s valid GLND representation for povm

sserita

Just bookkeeping: Not ready for review, rerequest from me when ready.

…e subset of models GLND, CPTPLND, H+S, S, full TP all to all when converting models with no noise. There is currently no support for converting to the same type for many of these

…ances

…re failing for some reason

juangmendoza19 · 2025-02-20T23:24:38Z

pygsti/baseobjs/errorgenbasis.py

+        TODO: 2 qubit labels returning None when label does exist in self.labels. 
+        '(support, left_support) not in self._offsets[eetype]' returning True incorrectly
+
+


The fancy code below (now commented out) to find the index of a label was not working for 2 qubits. Temporary easy solution implemented by just calling index(). This was creating problems during the creation of 2 qubit FOGI models

Pretty sure that I implemented a similar fix for this on PR #538. Making note of this in case there are merge conflicts that need to eventually get resolved here.

juangmendoza19 · 2025-02-20T23:27:11Z

pygsti/modelmembers/povms/__init__.py

+                            if ideal_effect is not None:
+                                base_items.append((lbl, convert_effect(ideal_effect, 'static', basis, ideal_effect, flatten_structure)))
+                            else:
+                                base_items.append((lbl, convert_effect(vec, 'static', basis, idl.get(lbl, None), flatten_structure)))


This is code to properly collect the ideal POVM effects. Before, the ideal_effect passed in was being ignored

juangmendoza19 · 2025-02-20T23:31:10Z

pygsti/modelmembers/povms/__init__.py

+                    raise ValueError("Failed to find an errorgen such that <ideal|exp(errorgen) = <effect|")
+                errgen_vec = _np.linalg.pinv(phys_directions)  @ soln.x
+                errorgen.from_vector(errgen_vec)
+


Code added to find errgen parameterization for POVMs based on optimization, just like state prep was being done before, but for some reason was never implemented on POVMs. Note that the optimization variations are done over a POVM space that lacks the dumb-gauge/metagauge. To find that space we wrote the function calc_physical_subspace which does a numerical calculation of the Jacobian with respect to POVM parameters.

juangmendoza19 · 2025-02-20T23:37:36Z

pygsti/modelmembers/states/__init__.py

+                    phys_directions = calc_physical_subspace(dense_state)
+
+                    #We use optimization to find the best error generator representation
+                    #we only vary physical directions, not independent error generators


Just as explained in the POVM init file, we added a function to identify the dumb-gauge/metagauge and move orthogonally to it. This allows for better optimization.

pygsti/modelmembers/states/__init__.py

juangmendoza19 · 2025-02-20T23:39:20Z

pygsti/models/explicitmodel.py

@@ -340,7 +340,7 @@ def __getitem__(self, label):
            raise KeyError("Key %s has an invalid prefix" % label)

    def convert_members_inplace(self, to_type, categories_to_convert='all', labels_to_convert='all',
-                                ideal_model=None, flatten_structure=False, set_default_gauge_group=False, cptp_truncation_tol= 1e-6):
+                                ideal_model=None, flatten_structure=False, set_default_gauge_group=False, cptp_truncation_tol= 1e-7, spam_cptp_penalty = 1e-6):


added piping for the cptp penalty factor to be used in the optimization within the SPAM convert functions discussed above

sserita

One refactor requested, and a few minor fixes while you are at it.

test/unit/objects/test_model.py

pygsti/modelmembers/povms/__init__.py

sserita

Forgot to hit Enter on my refactor request...

pygsti/modelmembers/states/__init__.py

coreyostrove · 2025-04-14T02:35:22Z

@juangmendoza19 and @sserita : Do either of you recall what still needs to get sorted out on this PR? I see that there is at least one merge conflict (probably from merging in develop recently) and some unresolved discussion threads above.

juangmendoza19 · 2025-04-14T17:00:22Z

@juangmendoza19 and @sserita : Do either of you recall what still needs to get sorted out on this PR? I see that there is at least one merge conflict (probably from merging in develop recently) and some unresolved discussion threads above.

Stefan’s requests above have not been implemented yet.

…ing issues for FOGI model construction

rileyjmurray

Left a few comments. Happy to iterate quickly to see them resolved!

pygsti/modelmembers/povms/__init__.py

pygsti/modelmembers/states/__init__.py

pygsti/tools/jamiolkowski.py

test/unit/objects/test_model.py

…smaller than this and perturbations along the S errgens showed up as the identity operator

pygsti/modelmembers/povms/__init__.py

rileyjmurray · 2025-05-27T04:08:25Z

@coreyostrove this looks good to me. Other than resolving any merge conflicts or failing tests, do you still have unaddressed comments?

I see that Stefan has unresolved comments. Rather than try to resolve those I suggest we dismiss Stefan's review (see screenshot below) and make a GitHub issue about following up on Stefan's unresolved points later.

test/unit/objects/test_output.txt

coreyostrove · 2025-05-27T04:17:02Z

@coreyostrove this looks good to me. Other than resolving any merge conflicts or failing tests, do you still have unaddressed comments?

I see that Stefan has unresolved comments. Rather than try to resolve those I suggest we dismiss Stefan's review (see screenshot below) and make a GitHub issue about following up on Stefan's unresolved points later.

One more minor change related to an errant file on my end, but after that I'm satisfied with all of this. I'll approve once that last change is made and we have confirmation that tests are still passing.

rileyjmurray · 2025-05-27T16:29:13Z

The failing tests seem related to the proposed changes:

FAILED test/unit/objects/test_model.py::TPModelTester::test_set_all_parameterizations_GLND - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
FAILED test/unit/objects/test_model.py::FullModelTester::test_set_all_parameterizations_GLND - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
FAILED test/unit/objects/test_model.py::StaticModelTester::test_set_all_parameterizations_GLND - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
FAILED test/unit/objects/test_model.py::LindbladModelTester::test_set_all_parameterizations_GLND - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
FAILED test/unit/drivers/test_longsequence.py::StdPracticeGSTTester::test_stdpractice_gst_CPTP - ValueError: Could not interpret 'CPTPLND' mode as a parameterization! Details:
Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
FAILED test/unit/objects/test_errorgenpropagation.py::ErrorgenPropTester::test_explicit_model - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
FAILED test/unit/drivers/test_longsequence.py::CPTPGatesTester::test_long_sequence_gst - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_bulk_fill_dprobs - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_bulk_fill_dprobs_with_block_size - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_bulk_fill_hprobs - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_bulk_fill_probs - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_doperation - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_hoperation - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
ERROR test/unit/objects/test_forwardsim.py::CPTPMatrixForwardSimTester::test_iter_hprobs_by_rectangle - ValueError: Failed to convert members. If converting to CPTP-based models, try providing an ideal_model to avoid possible branch cuts.
= 7 failed, 1934 passed, 64 skipped, 8103 warnings, 7 errors in 1757.83s (0:29:17) =
Error: Process completed with exit code 1.

Fixes are hopefully straightforward. @juangmendoza19, can you investigate?

Per the pyGSTi dev call today, we believe Juan has addressed all of Stefan's comments. All tests pass as of this moment, so I'm formally dismissing Stefan's review to remove the block from merging.

rileyjmurray · 2025-05-27T20:46:21Z

@coreyostrove all tests pass. If you approve then we can merge.

coreyostrove

Great work, @juangmendoza19! And thanks for bearing with the many changes requested during the review process.

juangmendoza19 added 6 commits July 16, 2024 14:59

povm convert fixed for fullTP->GLND, missing CPTPLND considerations

627ec20

full TP -> GLND reparam now avoids dumb-gauge, minimizes 2norm

484fe3e

bypass our changes if reparam is not full TP -> GLND

6365db8

removed magic numbers, adapted for arbitrary number of qubits

246623e

removed temp files

d4fd818

removed magic numbers from povm convert, and added a check if there i…

167d742

…s valid GLND representation for povm

juangmendoza19 requested review from rileyjmurray and a team as code owners August 23, 2024 18:12

juangmendoza19 requested a review from sserita August 23, 2024 18:12

juangmendoza19 added 2 commits September 4, 2024 09:22

commented povm file

c6ec19f

commented state file

2bf56d6

sserita reviewed Sep 19, 2024

View reviewed changes

juangmendoza19 added 11 commits September 26, 2024 11:58

started implementing CPTPLND compatibility with set_all_parameterization

0ec92dd

added support for converting to and from many other models. Tested th…

b769193

…e subset of models GLND, CPTPLND, H+S, S, full TP all to all when converting models with no noise. There is currently no support for converting to the same type for many of these

made cptp_penalty an optional argument, made it 1e-7 accross all inst…

e8232b3

…ances

Merge remote-tracking branch 'origin/develop' into bugfix-reparameterize

5425f61

using proper warning library, started unit tests

d70fe7d

unit test for process matrices in reparameterization added

51d75a5

fixed assert syntax

d39ea28

fixed small bug in errgen parameter counting, now forward sim tests a…

32970ff

…re failing for some reason

fixed warning bug in povm convert code

6786b14

added proper criteria to test parameterization change

933425b

added comment to remember to add more tests in the future

a737603

juangmendoza19 commented Feb 20, 2025

View reviewed changes

pygsti/modelmembers/states/__init__.py Outdated Show resolved Hide resolved

juangmendoza19 commented Feb 20, 2025

View reviewed changes

implemented Corey's requests for the PR

18b9294

juangmendoza19 requested a review from coreyostrove February 24, 2025 19:50

sserita requested changes Feb 27, 2025

View reviewed changes

test/unit/objects/test_model.py Outdated Show resolved Hide resolved

pygsti/modelmembers/povms/__init__.py Outdated Show resolved Hide resolved

sserita previously requested changes Feb 27, 2025

View reviewed changes

pygsti/modelmembers/states/__init__.py Outdated Show resolved Hide resolved

juangmendoza19 added 2 commits May 22, 2025 11:48

implemented Stefan's PR requests and cleaned up init.py s a little

482e86e

accepted label code from develop, although I believe it is still caus…

085188f

…ing issues for FOGI model construction

rileyjmurray requested changes May 22, 2025

View reviewed changes

juangmendoza19 added 7 commits May 23, 2025 11:09

applied many PR change requests made by Riley

d962130

final Riley PR suggestion regarding nontrivial singular vectors

e645650

fixed some parenthesis for multiplications in test file

a3d879f

removed debug print

5c1c42f

fixed problem with SV masking shape

d449c05

changed perturbation value that computes SPAM physical subspace. Any …

f94ce11

…smaller than this and perturbations along the S errgens showed up as the identity operator

cleaned up some leftover debugging code

9d7b4cf

rileyjmurray reviewed May 27, 2025

View reviewed changes

pygsti/modelmembers/povms/__init__.py Outdated Show resolved Hide resolved

compactified code for non trivial vectors

c90803d

rileyjmurray mentioned this pull request May 27, 2025

Audit use of eig versus eigh (and eigvals versus eigvalsh) #582

Open

rileyjmurray approved these changes May 27, 2025

View reviewed changes

coreyostrove reviewed May 27, 2025

View reviewed changes

test/unit/objects/test_output.txt Outdated Show resolved Hide resolved

removed null space vectors from svd to make dimensions match

9cb066c

coreyostrove approved these changes May 29, 2025

View reviewed changes

coreyostrove merged commit 3944ca5 into develop May 29, 2025
4 checks passed

coreyostrove deleted the bugfix-reparameterize branch May 29, 2025 00:59

		TODO: 2 qubit labels returning None when label does exist in self.labels.
		'(support, left_support) not in self._offsets[eetype]' returning True incorrectly

Bugfix reparameterize #480

Bugfix reparameterize #480

Uh oh!

Conversation

juangmendoza19 commented Aug 23, 2024

Uh oh!

sserita left a comment

Choose a reason for hiding this comment

Uh oh!

juangmendoza19 Feb 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

coreyostrove Feb 23, 2025

Choose a reason for hiding this comment

Uh oh!

juangmendoza19 Feb 20, 2025

Choose a reason for hiding this comment

Uh oh!

juangmendoza19 Feb 20, 2025

Choose a reason for hiding this comment

Uh oh!

juangmendoza19 Feb 20, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

juangmendoza19 Feb 20, 2025

Choose a reason for hiding this comment

Uh oh!

sserita left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

sserita left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

coreyostrove commented Apr 14, 2025

Uh oh!

juangmendoza19 commented Apr 14, 2025

Uh oh!

rileyjmurray left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

rileyjmurray commented May 27, 2025

Uh oh!

Uh oh!

coreyostrove commented May 27, 2025

Uh oh!

rileyjmurray commented May 27, 2025

Uh oh!

rileyjmurray commented May 27, 2025

Uh oh!

coreyostrove left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

juangmendoza19 Feb 20, 2025 •

edited

Loading