Regselect: catch errors in refit() (#77)

* catch error in refit() during regselect()

* test on Python 3.13

* allow Tikhonov regularizations to be None at initialization

* regselect: ensure ROM is fit() once correctly

* Hyeonghun pre-print

* small TimedBlock test coverage

* POD method of snapshots warning

* update version and changelog

Author: shanemcq18

.github/workflows/test_src.yml

@@ -26,6 +26,7 @@ jobs:
           - "3.10"
           - "3.11"
           - "3.12"
+          - "3.13"
     runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4

docs/literature.bib

@@ -656,3 +656,13 @@ @article{vijaywargiya2025tensoropinf
     doi = {10.48550/arXiv.2502.10888},
     category = {structure}
 }
+
+@article{kang2025semiconductor,
+    title = {Parametric {O}perator {I}nference to simulate the purging process in semiconductor manufacturing},
+    author = {Seunghyon Kang and Hyeonghun Kim and Boris Kramer},
+    journal = {arXiv},
+    volume = {2504.03990},
+    year = {2025},
+    doi = {10.48550/arXiv.2504.03990},
+    category = {application}
+}

docs/source/opinf/changelog.md

@@ -5,6 +5,13 @@
 New versions may introduce substantial new features or API adjustments.
 :::
 
+## Version 0.5.14
+
+- Catch any errors in `fit_regselect*()` that occur when the model uses `refit()`.
+- Tikhonov-type least-squares solvers do not require the regularizer in the constructor but will raise an `AttributeError` in `solve()` (and other methods) if the regularizer is not set. This makes using `fit_regselect_*()` much less cumbersome.
+- `PODBasis.fit(Q)` raises a warning when using the `"method-of-snapshots"`/`"eigh"` strategy if $n < k$ for $\mathbf{Q}\in\mathbb{R}^{n \times k}.$ In this case, calculating the $n \times k$ SVD is likely more efficient than the $k \times k$ eigenvalue problem.
+- Added Python 3.13 to list of tests.
+
 ## Version 0.5.13
 
 Bayesian operator inference:

pyproject.toml

@@ -39,13 +39,13 @@ dependencies = [
 [project.optional-dependencies]
 dev = [
     "bibtexparser>=2.0.0b7",
-    "tox>=4",
-    "pre-commit>=3.7.1",
-    "flake8==7.0.0",
     "black==24.4.2",
+    "flake8==7.0.0",
     "jupyterlab",
-    "pandas",
     "notebook",
+    "pandas",
+    "pre-commit>=3.7.1",
+    "tox>=4",
 ]
 
 [project.urls]

src/opinf/__init__.py

@@ -7,7 +7,7 @@
     https://github.com/Willcox-Research-Group/rom-operator-inference-Python3
 """
 
-__version__ = "0.5.13"
+__version__ = "0.5.14"
 
 from . import (
     basis,

src/opinf/basis/_pod.py

@@ -72,6 +72,7 @@ def method_of_snapshots(
     ----------
     states : (n, k) ndarray,
         Snapshot matrix :math:`\Q` from which to compute the POD vectors.
+        This method is computationally efficient when n >> k.
     inner_product_matrix : (n, n) sparse SPD matrix or None
         Spatial inner product matrix :math:`\W` for measuring how different
         indices in the snapshot matrix interact with each other.
@@ -100,7 +101,13 @@ def method_of_snapshots(
     any negative eigenvalues, then ``minthresh`` is increased to the absolute
     value of the most negative eigenvalue.
     """
-    n_states = states.shape[1]
+    state_dimension, n_states = states.shape
+    if n_states > state_dimension:
+        warnings.warn(
+            "state dimension < number of states, "
+            "method-of-snapshots / eigh may be inefficient",
+            errors.OpInfWarning,
+        )
     if inner_product_matrix is None:
         gramian = states.T @ (states / n_states)
     else:
@@ -113,7 +120,7 @@ def method_of_snapshots(
     eigvals = eigvals[::-1]
     eigvecs = eigvecs[:, ::-1]
 
-    # NOTE: By definition the Gramian is symmetric positive semi-definite.
+    # By definition the Gramian is symmetric positive semi-definite.
     # If any eigenvalues are smaller than zero, they are only measuring
     # numerical error and can be truncated.
     positives = eigvals > max(minthresh, abs(np.min(eigvals)))