Answer from cs61c-ep (Hao Wen Vivian Chong 15826235) for Question 3
Because floating-point addition was associative, and floating-point numbers have limited precision, better precision would be achieved by (0.5 - x) + 0.5 than if one used 1.0 - x.