I have found a workaround, using tt.set_subtensor() in a loop to assign output values to an empty array. Painfully slow but it worked.
I have found a workaround, using tt.set_subtensor() in a loop to assign output values to an empty array. Painfully slow but it worked.