Fix gradient of minimize and root wrt higher-dimensional arguments (#1806)

* More precise imports

* Fix gradient of `minimize` and `root` wrt higher-dimensional arguments

Also:
* Allow several inputs to be optimized
* Handle disconnected and undefined gradients

---------

Co-authored-by: ricardoV94 <[email protected]>

Author: jessegrabowski

pytensor/sparse/math.py

@@ -12,8 +12,8 @@
 from pytensor.gradient import grad_not_implemented
 from pytensor.graph import Apply, Op
 from pytensor.link.c.op import COp
-from pytensor.tensor import TensorType, Variable, specify_broadcastable, tensor
-from pytensor.tensor.type import complex_dtypes
+from pytensor.tensor.shape import specify_broadcastable
+from pytensor.tensor.type import TensorType, Variable, complex_dtypes, tensor
 
 
 def structured_elemwise(tensor_op):

pytensor/sparse/variable.py

@@ -30,8 +30,8 @@
 )
 from pytensor.sparse.type import SparseTensorType
 from pytensor.sparse.utils import hash_from_sparse
-from pytensor.tensor import iscalar
 from pytensor.tensor.shape import shape
+from pytensor.tensor.type import iscalar
 from pytensor.tensor.variable import (
     TensorConstant,
     TensorVariable,