We have seen that using an integer to approximate a real (floating-point) quantity causes an error. If we record the error as it accumulates, we can use this information to choose the pixels which best approximate the line.