There are only so may unique floating point numbers that fall into that range. Say I scale the data up to [0, FLT_MAX]-- the finite number of possible floating point values has increased dramatically.