start gradient tests

Author: jessegrabowski

tests/link/numba/test_slinalg.py

@@ -7,6 +7,7 @@
 import pytensor
 import pytensor.tensor as pt
 from pytensor.graph import FunctionGraph
+from tests import unittest_tools as utt
 from tests.link.numba.test_basic import compare_numba_and_py
 
 
@@ -157,6 +158,16 @@ def test_numba_Cholesky_raise_on(on_error):
         assert np.all(np.isnan(f(test_value)))
 
 
+@pytest.mark.parametrize("lower", [True, False], ids=["lower=True", "lower=False"])
+@pytest.mark.parametrize("trans", [True, False], ids=["trans=True", "trans=False"])
+def test_numba_Cholesky_grad(lower, trans):
+    rng = np.random.default_rng(utt.fetch_seed())
+    L = rng.random.normal(size=(5, 5)).astype(floatX)
+    X = L @ L.T
+
+    utt.verify_grad(pt.linalg.cholesky, [X])
+
+
 def test_block_diag():
     A = pt.matrix("A")
     B = pt.matrix("B")
@@ -358,6 +369,8 @@ def test_solve(b_func, b_size, assume_a, transposed):
     assert (A_val == A_val_copy).all() == (op.destroy_map.get(0, None) != [0])
     assert (b_val == b_val_copy).all() == (op.destroy_map.get(0, None) != [1])
 
+    # Test gradients
+
 
 @pytest.mark.parametrize(
     "b_func, b_size",