fix(numba): cholesky did not set off-diag entries to zero

Author: aseyboldt

pytensor/link/numba/dispatch/slinalg.py

@@ -346,6 +346,15 @@ def impl(A, lower=0, overwrite_a=False, check_finite=True):
             INFO,
         )
 
+        if lower:
+            for j in range(1, _N):
+                for i in range(j):
+                    A_copy[i, j] = 0.0
+        else:
+            for j in range(_N):
+                for i in range(j + 1, _N):
+                    A_copy[i, j] = 0.0
+
         return A_copy, int_ptr_to_val(INFO)
 
     return impl

tests/link/numba/test_slinalg.py

@@ -2,14 +2,13 @@
 
 import numpy as np
 import pytest
+import scipy
 
 import pytensor
 import pytensor.tensor as pt
 from pytensor import config
-from pytensor.compile import SharedVariable
-from pytensor.graph import Constant, FunctionGraph
+from pytensor.graph import FunctionGraph
 from tests.link.numba.test_basic import compare_numba_and_py
-from tests.tensor.test_extra_ops import set_test_value
 
 
 numba = pytest.importorskip("numba")
@@ -109,23 +108,25 @@ def test_solve_triangular_raises_on_nan_inf(value):
 
 
 @pytest.mark.parametrize("lower", [True, False], ids=["lower=True", "lower=False"])
-def test_numba_Cholesky(lower):
-    x = set_test_value(
-        pt.tensor(dtype=config.floatX, shape=(3, 3)),
-        (lambda x: x.T.dot(x))(rng.random(size=(3, 3)).astype(config.floatX)),
-    )
+@pytest.mark.parametrize("trans", [True, False], ids=["trans=True", "trans=False"])
+def test_numba_Cholesky(lower, trans):
+    cov = pt.matrix("cov")
 
-    g = pt.linalg.cholesky(x, lower=lower)
-    g_fg = FunctionGraph(outputs=[g])
+    if trans:
+        cov_ = cov.T
+    else:
+        cov_ = cov
+    chol = pt.linalg.cholesky(cov_, lower=lower)
 
-    compare_numba_and_py(
-        g_fg,
-        [
-            i.tag.test_value
-            for i in g_fg.inputs
-            if not isinstance(i, SharedVariable | Constant)
-        ],
-    )
+    fg = FunctionGraph(outputs=[chol])
+
+    x = np.array([0.1, 0.2, 0.3])
+    val = np.eye(3) + x[None, :] + x[:, None]
+
+    compare_numba_and_py(fg, [val])
+
+    func = pytensor.function([cov], chol, mode="NUMBA")
+    np.testing.assert_allclose(func(val), scipy.linalg.cholesky(val, lower=lower))
 
 
 def test_numba_Cholesky_raises_on_nan_input():