Fix vectorized shape

Author: ricardoV94

pytensor/tensor/shape.py

@@ -8,7 +8,7 @@
 import pytensor
 from pytensor.gradient import DisconnectedType
 from pytensor.graph.basic import Apply, Variable
-from pytensor.graph.replace import _vectorize_node, _vectorize_not_needed
+from pytensor.graph.replace import _vectorize_node
 from pytensor.graph.type import HasShape
 from pytensor.link.c.op import COp
 from pytensor.link.c.params_type import ParamsType
@@ -155,7 +155,20 @@ def _get_vector_length_Shape(op, var):
     return var.owner.inputs[0].type.ndim
 
 
-_vectorize_node.register(Shape, _vectorize_not_needed)
+@_vectorize_node.register(Shape)
+def vectorize_shape(op, node, batched_x):
+    from pytensor.tensor.extra_ops import broadcast_to
+
+    [old_x] = node.inputs
+    core_ndims = old_x.type.ndim
+    batch_ndims = batched_x.type.ndim - core_ndims
+    batched_x_shape = shape(batched_x)
+    if not batch_ndims:
+        return batched_x_shape.owner
+    else:
+        batch_shape = batched_x_shape[:batch_ndims]
+        core_shape = batched_x_shape[batch_ndims:]
+        return broadcast_to(core_shape, (*batch_shape, core_ndims)).owner
 
 
 def shape_tuple(x: TensorVariable) -> tuple[Variable, ...]:

tests/tensor/test_shape.py

@@ -709,29 +709,66 @@ def test_shape_tuple():
 
 
 class TestVectorize:
+    @pytensor.config.change_flags(cxx="")  # For faster eval
     def test_shape(self):
-        vec = tensor(shape=(None,))
-        mat = tensor(shape=(None, None))
-
+        vec = tensor(shape=(None,), dtype="float64")
+        mat = tensor(shape=(None, None), dtype="float64")
         node = shape(vec).owner
-        vect_node = vectorize_node(node, mat)
-        assert equal_computations(vect_node.outputs, [shape(mat)])
 
+        [vect_out] = vectorize_node(node, mat).outputs
+        assert equal_computations(
+            [vect_out], [broadcast_to(mat.shape[1:], (*mat.shape[:1], 1))]
+        )
+
+        mat_test_value = np.ones((5, 3))
+        ref_fn = np.vectorize(lambda vec: np.asarray(vec.shape), signature="(vec)->(1)")
+        np.testing.assert_array_equal(
+            vect_out.eval({mat: mat_test_value}),
+            ref_fn(mat_test_value),
+        )
+
+        mat = tensor(shape=(None, None), dtype="float64")
+        tns = tensor(shape=(None, None, None, None), dtype="float64")
+        node = shape(mat).owner
+        [vect_out] = vectorize_node(node, tns).outputs
+        assert equal_computations(
+            [vect_out], [broadcast_to(tns.shape[2:], (*tns.shape[:2], 2))]
+        )
+
+        tns_test_value = np.ones((4, 6, 5, 3))
+        ref_fn = np.vectorize(
+            lambda vec: np.asarray(vec.shape), signature="(m1,m2)->(2)"
+        )
+        np.testing.assert_array_equal(
+            vect_out.eval({tns: tns_test_value}),
+            ref_fn(tns_test_value),
+        )
+
+    @pytensor.config.change_flags(cxx="")  # For faster eval
     def test_reshape(self):
         x = scalar("x", dtype=int)
-        vec = tensor(shape=(None,))
-        mat = tensor(shape=(None, None))
+        vec = tensor(shape=(None,), dtype="float64")
+        mat = tensor(shape=(None, None), dtype="float64")
 
-        shape = (2, x)
+        shape = (-1, x)
         node = reshape(vec, shape).owner
-        vect_node = vectorize_node(node, mat, shape)
-        assert equal_computations(
-            vect_node.outputs, [reshape(mat, (*mat.shape[:1], 2, x))]
+
+        [vect_out] = vectorize_node(node, mat, shape).outputs
+        assert equal_computations([vect_out], [reshape(mat, (*mat.shape[:1], -1, x))])
+
+        x_test_value = 2
+        mat_test_value = np.ones((5, 6))
+        ref_fn = np.vectorize(
+            lambda x, vec: vec.reshape(-1, x), signature="(),(vec1)->(mat1,mat2)"
+        )
+        np.testing.assert_array_equal(
+            vect_out.eval({x: x_test_value, mat: mat_test_value}),
+            ref_fn(x_test_value, mat_test_value),
         )
 
-        new_shape = (5, 2, x)
-        vect_node = vectorize_node(node, mat, new_shape)
-        assert equal_computations(vect_node.outputs, [reshape(mat, new_shape)])
+        new_shape = (5, -1, x)
+        [vect_out] = vectorize_node(node, mat, new_shape).outputs
+        assert equal_computations([vect_out], [reshape(mat, new_shape)])
 
         with pytest.raises(NotImplementedError):
             vectorize_node(node, vec, broadcast_to(as_tensor([5, 2, x]), (2, 3)))