Hey, I want to create a nxn
tensor with a repeating vector starting on the diagonal and the rest initialized with zeros. For Example:
vector v = [v_1, v_2, v_3]
v_1, v_2, v_3, 0
0, v_1, v_2, v_3
0, 0, v_1, v_2
0, 0, 0, v_1
Is this possible? It should be, but I am currently not sure how…also can whatever the solution might be: is it possible to backprob through it?