Use lapack func instead of `scipy.linalg.cholesky` (#1487)

pytensor/tensor/slinalg.py

@@ -37,7 +37,7 @@ class Cholesky(Op):
         self,
         *,
         lower: bool = True,
-        check_finite: bool = True,
+        check_finite: bool = False,
         on_error: Literal["raise", "nan"] = "raise",
         overwrite_a: bool = False,
     ):
@@ -67,29 +67,55 @@ class Cholesky(Op):
     def perform(self, node, inputs, outputs):
         [x] = inputs
         [out] = outputs
-        try:
-            # Scipy cholesky only makes use of overwrite_a when it is F_CONTIGUOUS
-            # If we have a `C_CONTIGUOUS` array we transpose to benefit from it
-            if self.overwrite_a and x.flags["C_CONTIGUOUS"]:
-                out[0] = scipy_linalg.cholesky(
-                    x.T,
-                    lower=not self.lower,
-                    check_finite=self.check_finite,
-                    overwrite_a=True,
-                ).T
-            else:
-                out[0] = scipy_linalg.cholesky(
-                    x,
-                    lower=self.lower,
-                    check_finite=self.check_finite,
-                    overwrite_a=self.overwrite_a,
-                )
 
-        except scipy_linalg.LinAlgError:
-            if self.on_error == "raise":
-                raise
+        (potrf,) = scipy_linalg.get_lapack_funcs(("potrf",), (x,))
+
+        # Quick return for square empty array
+        if x.size == 0:
+            out[0] = np.empty_like(x, dtype=potrf.dtype)
+            return
+
+        if self.check_finite and not np.isfinite(x).all():
+            if self.on_error == "nan":
+                out[0] = np.full(x.shape, np.nan, dtype=potrf.dtype)
+                return
             else:
+                raise ValueError("array must not contain infs or NaNs")
+
+        # Squareness check
+        if x.shape[0] != x.shape[1]:
+            raise ValueError(
+                "Input array is expected to be square but has " f"the shape: {x.shape}."
+            )
+
+        # Scipy cholesky only makes use of overwrite_a when it is F_CONTIGUOUS
+        # If we have a `C_CONTIGUOUS` array we transpose to benefit from it
+        c_contiguous_input = self.overwrite_a and x.flags["C_CONTIGUOUS"]
+        if c_contiguous_input:
+            x = x.T
+            lower = not self.lower
+            overwrite_a = True
+        else:
+            lower = self.lower
+            overwrite_a = self.overwrite_a
+
+        c, info = potrf(x, lower=lower, overwrite_a=overwrite_a, clean=True)
+
+        if info != 0:
+            if self.on_error == "nan":
                 out[0] = np.full(x.shape, np.nan, dtype=node.outputs[0].type.dtype)
+            elif info > 0:
+                raise scipy_linalg.LinAlgError(
+                    f"{info}-th leading minor of the array is not positive definite"
+                )
+            elif info < 0:
+                raise ValueError(
+                    f"LAPACK reported an illegal value in {-info}-th argument "
+                    f'on entry to "POTRF".'
+                )
+        else:
+            # Transpose result if input was transposed
+            out[0] = c.T if c_contiguous_input else c
 
     def L_op(self, inputs, outputs, gradients):
         """
@@ -201,7 +227,9 @@ def cholesky(
 
     """
 
-    return Blockwise(Cholesky(lower=lower, on_error=on_error))(x)
+    return Blockwise(
+        Cholesky(lower=lower, on_error=on_error, check_finite=check_finite)
+    )(x)
 
 
 class SolveBase(Op):

tests/link/numba/test_slinalg.py

@@ -465,7 +465,7 @@ def test_cholesky_raises_on_nan_input():
 
     x = pt.tensor(dtype=floatX, shape=(3, 3))
     x = x.T.dot(x)
-    g = pt.linalg.cholesky(x)
+    g = pt.linalg.cholesky(x, check_finite=True)
     f = pytensor.function([x], g, mode="NUMBA")
 
     with pytest.raises(np.linalg.LinAlgError, match=r"Non-numeric values"):

tests/tensor/test_slinalg.py

@@ -74,6 +74,26 @@ def test_cholesky():
     check_upper_triangular(pd, ch_f)
 
 
+def test_cholesky_performance(benchmark):
+    rng = np.random.default_rng(utt.fetch_seed())
+    r = rng.standard_normal((10, 10)).astype(config.floatX)
+    pd = np.dot(r, r.T)
+    x = matrix()
+    chol = cholesky(x)
+    ch_f = function([x], chol)
+    benchmark(ch_f, pd)
+
+
+def test_cholesky_empty():
+    empty = np.empty([0, 0], dtype=config.floatX)
+    x = matrix()
+    chol = cholesky(x)
+    ch_f = function([x], chol)
+    ch = ch_f(empty)
+    assert ch.size == 0
+    assert ch.dtype == config.floatX
+
+
 def test_cholesky_indef():
     x = matrix()
     mat = np.array([[1, 0.2], [0.2, -2]]).astype(config.floatX)