Fix too strict type check in _sum_grad_over_bcasted_dims

pytensor/tensor/subtensor.py

@@ -2027,7 +2027,6 @@ def _sum_grad_over_bcasted_dims(x, gx):
     if gx.broadcastable != x.broadcastable:
         x_dim_added = gx.ndim - x.ndim
         x_broad = (True,) * x_dim_added + x.broadcastable
-        assert sum(gx.broadcastable) <= sum(x_broad)
         axis_to_sum = []
         for i in range(gx.ndim):
             if gx.broadcastable[i] is False and x_broad[i] is True:
@@ -2045,7 +2044,14 @@ def _sum_grad_over_bcasted_dims(x, gx):
             for i in range(x_dim_added):
                 assert gx.broadcastable[i]
             gx = gx.dimshuffle(*range(x_dim_added, gx.ndim))
-        assert gx.broadcastable == x.broadcastable
+        # Broadcastable flags of gx can be the same or more specific than x.
+        # Only unallowed case is x_dim_b == True and gx_dim_b == False.
+        assert not any(
+            x_dim_b and not gx_dim_b
+            for x_dim_b, gx_dim_b in zip(
+                x.type.broadcastable, gx.type.broadcastable, strict=True
+            )
+        ), (x.type, gx.type)
     return gx
 
 

tests/tensor/test_subtensor.py

@@ -12,7 +12,9 @@
 from pytensor import function
 from pytensor.compile import DeepCopyOp, shared
 from pytensor.compile.io import In
+from pytensor.compile.mode import Mode
 from pytensor.configdefaults import config
+from pytensor.gradient import grad
 from pytensor.graph.op import get_test_value
 from pytensor.graph.rewriting.utils import is_same_graph
 from pytensor.printing import pprint
@@ -22,6 +24,7 @@
 from pytensor.tensor.elemwise import DimShuffle
 from pytensor.tensor.math import exp, isinf
 from pytensor.tensor.math import sum as pt_sum
+from pytensor.tensor.shape import specify_shape
 from pytensor.tensor.subtensor import (
     AdvancedIncSubtensor,
     AdvancedIncSubtensor1,
@@ -1660,6 +1663,25 @@ def just_numeric_args(a, b):
                 ),
             )
 
+    def test_grad_broadcastable_specialization(self):
+        # Make sure gradient does not fail when gx has a more precise static_shape after indexing.
+        # This is a regression test for a bug reported in
+        # https://discourse.pymc.io/t/marginalized-mixture-wont-begin-sampling-throws-assertion-error/15969
+
+        x = vector("x")  # Unknown write time shape = (2,)
+        out = x.zeros_like()
+
+        # Update a subtensor of unknown write time shape = (1,)
+        out = out[1:].set(exp(x[1:]))
+        out = specify_shape(out, 2)
+        gx = grad(out.sum(), x)
+
+        mode = Mode(linker="py", optimizer=None)
+        np.testing.assert_allclose(
+            gx.eval({x: [1, 1]}, mode=mode),
+            [0, np.e],
+        )
+
 
 class TestIncSubtensor1:
     def setup_method(self):