Closed
Description
Description
The following graph is invalid:
x = pt.tensor('x', shape=(5,3)
y = pt.linalg.det(x)
This will happily construct a graph, but it will always fail at runtime when the perform
method is called, because the numpy implementation will raise an error. There's no reason we can't check add a check to make_node
testing if 1) the shape of the input is known and, 2) if it is valid.