Only copy numpy arrays when necessary

.github/workflows/build-wheels.yml

@@ -17,7 +17,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-22.04, ubuntu-22.04-arm, windows-2022, macos-13, macos-14]
+        os: [ubuntu-22.04, ubuntu-22.04-arm, windows-2022, macos-14, macos-15]
 
     steps:
       - uses: actions/checkout@1af3b93b6815bc44a9784bd300feb67ff0d1eeb3 # v6.0.0

CHANGELOG.rst

@@ -8,6 +8,15 @@ Changelog
 =========
 
 
+4.1.4 - 2025-12-16
+------------------
+
+**Other changes:**
+
+- :func:`tabmat.from_df` now avoids unnecessary copies of dense arrays, but still ensures that the results are contiguous (C or F order).
+- We now use `narwhals`' v2 API for data frame handling.
+
+
 4.1.3 - 2025-11-12
 ------------------
 

src/tabmat/categorical_matrix.py

@@ -509,7 +509,7 @@ def matvec(
         is_int = np.issubdtype(other.dtype, np.signedinteger)
 
         if is_int:
-            other_m = other.astype(float)  # type: ignore
+            other_m: np.ndarray = other.astype(float)
         else:
             other_m = other
 

src/tabmat/constructor.py

@@ -182,7 +182,7 @@ def from_df(
     if dense_dfidx:
         matrices.append(
             DenseMatrix(
-                df[:, dense_dfidx].to_numpy().astype(dtype),
+                df[:, dense_dfidx].to_numpy().astype(dtype, copy=False),
                 column_names=np.asarray(df.columns)[dense_dfidx],
                 term_names=np.asarray(df.columns)[dense_dfidx],
             )

src/tabmat/dense_matrix.py

@@ -1,4 +1,5 @@
 import textwrap
+import warnings
 from typing import Optional, Union
 
 import numpy as np
@@ -43,6 +44,19 @@ def __init__(self, input_array, column_names=None, term_names=None):
         elif input_array.ndim > 2:
             raise ValueError("Input array must be 1- or 2-dimensional")
 
+        # Ensure array is contiguous (C or F order) for Cython operations
+        # Only copy if necessary
+        if (
+            not input_array.flags["C_CONTIGUOUS"]
+            and not input_array.flags["F_CONTIGUOUS"]
+        ):
+            warnings.warn(
+                "Input array is not contiguous; making a copy.",
+                UserWarning,
+                stacklevel=2,
+            )
+            input_array = np.asfortranarray(input_array)
+
         self._array = input_array
         width = self._array.shape[1]
 

src/tabmat/formula.py

@@ -92,7 +92,7 @@ def _encode_numerical(self, values, metadata, encoder_state, spec, drop_rows):
         if drop_rows:
             values = values.drop(index=values.index[drop_rows])
         if isinstance(values, pd.Series):
-            values = values.to_numpy().astype(self.dtype)
+            values = values.to_numpy().astype(self.dtype, copy=False)
         if (values != 0).mean() <= self.sparse_threshold:
             return _InteractableSparseVector(sps.csc_matrix(values[:, np.newaxis]))
         else:

tests/test_fast_sandwich.py

@@ -64,25 +64,30 @@ def test_fast_sandwich_dense():
 
 
 def test_dense_sandwich_on_non_contiguous():
-    """Non-regression test for #208"""
+    """Non-regression test for #208
+
+    DenseMatrix now automatically ensures arrays are contiguous,
+    so non-contiguous inputs are automatically copied and made contiguous.
+    """
     rng = np.random.default_rng(seed=123)
     X = rng.standard_normal(size=(100, 20))
 
-    # Xd wraps a not-contiguous array.
-    Xd = DenseMatrix(X[:, :10])
-    Xs = SparseMatrix(csc_matrix(X[:, 10:]))
-    Xm = SplitMatrix([Xd, Xs])
+    # Column slicing creates a non-contiguous array, but DenseMatrix
+    # automatically makes it contiguous (copying only if necessary).
+    non_contiguous_array = X[:, :10]
+    assert not non_contiguous_array.flags["C_CONTIGUOUS"]
+    assert not non_contiguous_array.flags["F_CONTIGUOUS"]
 
-    # Making the sandwich product fail.
-    with pytest.raises(Exception, match="The matrix X is not contiguous"):
-        Xm.sandwich(np.ones(X.shape[0]))
+    Xd = DenseMatrix(non_contiguous_array)
+    # The internal array should now be contiguous
+    assert Xd.A.flags["C_CONTIGUOUS"] or Xd.A.flags["F_CONTIGUOUS"]
 
-    # Xd wraps a copy, which makes the data contiguous.
-    Xd = DenseMatrix(X[:, :10].copy())
+    Xs = SparseMatrix(csc_matrix(X[:, 10:]))
     Xm = SplitMatrix([Xd, Xs])
 
-    # The sandwich product works without problem here.
-    Xm.sandwich(np.ones(X.shape[0]))
+    # The sandwich product should work without problem
+    result = Xm.sandwich(np.ones(X.shape[0]))
+    assert result is not None
 
 
 def check(A, d, cols):