FOGI model serialization and some FOGI model optimization #598

juangmendoza19 · 2025-06-24T21:13:38Z

This resolves issue #595 by allowing FOGI models to be serialized. Now, any fogi model can be saved to disk with the method .write(''), and loaded back with pygsti.models.Model.read('').

The main problem was the the class FirstOrderGaugeInvariantStore was not nicely serializable, which is stored in FOGI models. All appropriate serializations were added to this class, and other classes used within.

Additionally, @coreyostrove and I modified model.py:set_parameter_values() to be executed faster when used with FOGI models

juangmendoza19 · 2025-06-24T21:14:32Z

pygsti/baseobjs/errorgenbasis.py

@@ -88,6 +89,7 @@ def __init__(self, state_space, labels, basis_1q=None):
            comprise the basis element labels for the values of the
            `ElementaryErrorgenLabels` in `labels`.
        """
+        super().__init__()


Every object that is nicely serializable needs to have this init statement

juangmendoza19 · 2025-06-24T21:18:57Z

pygsti/baseobjs/errorgenbasis.py

+        return state
+    @classmethod
+    def from_nice_serialization(cls, state):
+        return cls(_StateSpace.from_nice_serialization(state['state_space']), [_GlobalElementaryErrorgenLabel.cast(label) for label in state['labels']], state['_basis_1q'] if isinstance(state['_basis_1q'], str) else _Basis.from_nice_serialization(state['_basis_1q']))


Added support for _basis_1q variables which can be strings or Basis objects according to the docstring

juangmendoza19 · 2025-06-24T21:19:49Z

pygsti/baseobjs/statespace.py

@@ -418,7 +418,7 @@ def create_subspace(self, labels):
        StateSpace
        """
        # Default, generic, implementation constructs an explicit state space
-        labels = set(labels)
+        labels = sorted(set(labels))


This change was made with @coreyostrove to make FOGI model construction deterministic

juangmendoza19 · 2025-06-24T21:21:35Z

pygsti/models/fogistore.py

+        self.gauge_space = gauge_space
+        self.elem_errorgen_labels_by_op = elem_errorgen_labels_by_op
+        self.op_errorgen_indices = op_errorgen_indices
+        self.fogi_directions = fogi_directions


The init method of this class used to not only create a FOGIstore object, but also compute all fogi directions. In order to be able to load them from disk, we needed a simple init method which just takes attributes and spits back the object without doing any math.

juangmendoza19 · 2025-06-24T21:22:56Z

pygsti/models/fogistore.py

+        self.fogv_labels = ["%d gauge action" % i for i in range(self.fogv_directions.shape[1])]
+
+    @classmethod
+    def compute_fogis(cls, gauge_action_matrices_by_op, gauge_action_gauge_spaces_by_op, errorgen_coefficient_labels_by_op,


The old __init__method became compute_fogis(). It is decorated as a classmethod such that it can be used without the existence of a fogi_store object. As it intended to be used, at the end it creates an object which is returned back to the user.

pygsti/models/fogistore.py

juangmendoza19 · 2025-06-24T21:25:29Z

pygsti/models/model.py

@@ -657,56 +657,6 @@ def num_modeltest_params(self):
        self._clean_paramvec()
        return Model.num_modeltest_params.fget(self)

-    @property


It seems like a mistake that develop had all these methods duplicated

juangmendoza19 · 2025-06-24T21:26:40Z

pygsti/models/model.py

+                index_mm_map[gpidx].append(obj)
+                index_mm_label_map[gpidx].append(lbl)
+        self._index_mm_map = index_mm_map
+        self._index_mm_label_map = index_mm_label_map


the resulting index_mm_map was incorrect for FOGI models. This is necessary for our set_parameter_values optimization

juangmendoza19 · 2025-06-24T21:29:27Z

pygsti/models/model.py

+
+            indices = non_zero_errgens[0]
+            values = new_errgen_vec[indices]
+            vec_to_access = new_errgen_vec


This if statement captures which error generators are affected by the FOGI vector changing.

vec_to_access is a new variable which will always hold the vector of object parameters. In the case of FOGI models, it will always correspond to the updated error generator vector.

juangmendoza19 · 2025-06-24T21:30:50Z

pygsti/models/model.py

+        #and c) the effect is a child of that POVM.
+
+        if isinstance(self._layer_rules, _ExplicitLayerRules):
+            updated_children = []


unique_mms identifies which operations correspond to the error generators that are changing, which then allows us to only update those in our model inside the loop. This way we only update the objects in our model which are affected by the values that were modified by the user.

juangmendoza19 · 2025-06-24T21:37:59Z

pygsti/models/model.py

@@ -1304,52 +1253,71 @@ def set_parameter_values(self, indices, values, close=False):
        -------
        None
        """
+        orig_param_vec = self._paramvec.copy()


All of the code below assumes that parameter interposers are only used for FOGI models, and that FOGI models are built from GLND parameterizations (i.e. every FOGI quantity is defined in terms of error generators)

pygsti/models/model.py

coreyostrove

Mostly looks good. A few requests inline. Also a general request to identify whether or not we have unit test coverage for the new functionality that is being added for both the serialization cases and the updated code path for set_parameter_values when using a FOGI model/model with interposer.

pygsti/baseobjs/errorgenbasis.py

pygsti/baseobjs/errorgenspace.py

pygsti/models/fogistore.py

… local errgenbasis labels

…ks but does not return his result

coreyostrove

Only a few very minor requests for a) a couple one-sentence docstrings (see comments in code) and b) for another merge of develop into your branch to sync up changes from PRs merged in earlier today (there is at least one conflict/unintentional reversion that would be introduced otherwise). After that I think this is good to go.

pygsti/baseobjs/errorgenbasis.py

pygsti/baseobjs/errorgenspace.py

pygsti/models/fogistore.py

pygsti/models/modelparaminterposer.py

test/unit/objects/test_model.py

juangmendoza19 · 2025-07-14T17:45:01Z

test/unit/objects/test_model.py

@@ -730,3 +746,55 @@ def tearDown(self):
 #    def test_randomize_with_unitary_raises(self):
 #        with self.assertRaises(AssertionError):
 #            self.model.randomize_with_unitary(1, rand_state=np.random.RandomState())  # scale shouldn't matter
+


All of the code below is not original to this PR, it was moved from test_nice_serialization.py

coreyostrove · 2025-07-28T16:49:47Z

pygsti/tools/fogitools.py

+
+                            # Step 2: Construct Sigma
+                            Sigma = _np.zeros((U.shape[0], VT.shape[0]))
+                            _np.fill_diagonal(Sigma, S)


Pro-tip, use broadcasting here. It is much more efficient. This Sigma matrix is only being constructed here for the sake of the matrix multiplication on line 578, but what does matrix multiplication by a diagonal matrix, D, and some generic matrix A do? It rescales each row of A by the corresponding diagonal entry of D. If you have a vector V of the correct size (len=num_rows(A)) and you multiply it into A (using the standard multiplication operator, '*', not matrix multiplication's '@') then numpy's default behavior is to broadcast this multiplication across A, which is hard to describe concisely, but has exactly the desired effect of rescaling the rows of A by the corresponding values of V. But you'll achieve this with better computational complexity (only a quadratic number of multiplications instead of the full arithmetic cost of matrix multiplication).

This isn't a bottleneck as far as I am aware, so I'm not going to hold things up for the sake of this change, but there's never a bad time to learn about nifty numpy performance tricks.

Correct the kwarg on qr decomposition.

coreyostrove

Tests are now passing, which was the last thing I was waiting on before approval. Great work, @juangmendoza19!

…into feature-ams-2

juangmendoza19 added 3 commits June 24, 2025 14:28

recovered mostly changes to allow for FOGI model serialization

106ecbe

changed back some changes from develop that were lost

f13821d

small white space corrections

67f31f7

juangmendoza19 requested review from a team as code owners June 24, 2025 21:13

juangmendoza19 requested a review from coreyostrove June 24, 2025 21:13

juangmendoza19 commented Jun 24, 2025

View reviewed changes

pygsti/models/fogistore.py Show resolved Hide resolved

juangmendoza19 commented Jun 24, 2025

View reviewed changes

pygsti/models/model.py Show resolved Hide resolved

juangmendoza19 added 2 commits June 24, 2025 16:05

fixed fogistore attribute name

f358b32

fixed other instances of dependent_fogi_action being incorrectly named

8fc9fa2

coreyostrove added this to the 0.9.14 milestone Jun 25, 2025

coreyostrove requested changes Jun 25, 2025

View reviewed changes

juangmendoza19 added 7 commits June 25, 2025 17:13

added serialization methods for completerrgenbasis, added support for…

b15ff9b

… local errgenbasis labels

removed print statements

5b4679e

started docstring

dd679e8

finished fogistore docstring

8b2f676

implemented and tested __eq__ method for fogistore

2a46692

added test for FOGI equal method

6703ab8

added tests for fogi serialization

bc39b12

juangmendoza19 requested a review from rileyjmurray as a code owner June 27, 2025 17:30

juangmendoza19 added 4 commits June 27, 2025 13:00

added set_param_values unit tests for param_interposer/fogi models

f40ee35

added a docstring and fixed vector decoding for PR

acf96f1

fixed indentation

17dab75

finished docstring for FOGIStore

3d5b33d

juangmendoza19 requested a review from coreyostrove June 30, 2025 21:26

juangmendoza19 added 7 commits June 30, 2025 16:17

fixed issue with FOGIStore method calling

6ffdf9e

fixed fogistore

b6fde07

small patch on label_index problem that has been identified before

4f2ce40

created amalgamation of Erik's label_index and mine, it uses his chec…

2f00ce8

…ks but does not return his result

removed debug print

ad2526c

finally properly fixed FOGI label_index bug

ca32580

moved unit tests to correct file

f5aa81e

coreyostrove requested changes Jul 9, 2025

View reviewed changes

juangmendoza19 added 8 commits July 9, 2025 10:41

added __eq__ docstring

3f66504

added another dosctring

965c785

added another docstring

b1078dc

removed whitespace

e392502

improved label_index tests

67deafd

possible fix for when svd does not converge

6784b87

Merge branch 'develop' into feature-ams-2

001cbce

fixed import

2df848a

juangmendoza19 commented Jul 14, 2025

View reviewed changes

juangmendoza19 added 2 commits July 16, 2025 15:43

removed commented out code

3b401b2

1000x Murray update

f17d24a

coreyostrove reviewed Jul 28, 2025

View reviewed changes

Fix kwarg

2171e3b

Correct the kwarg on qr decomposition.

coreyostrove approved these changes Jul 29, 2025

View reviewed changes

juangmendoza19 added 2 commits August 4, 2025 10:27

removed unnecessary diagonal matrix as pointed out by Corey

b68f182

Merge branch 'feature-ams-2' of https://github.com/sandialabs/pyGSTi …

18043b4

…into feature-ams-2

coreyostrove merged commit b599d9e into develop Aug 4, 2025
4 checks passed

FOGI model serialization and some FOGI model optimization #598

FOGI model serialization and some FOGI model optimization #598

Uh oh!

Conversation

juangmendoza19 commented Jun 24, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

coreyostrove left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

coreyostrove left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

coreyostrove left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!