His final solution seems to be very similar in motivation to fixed point arithemetic, but with a “floating” point. I wonder how those approaches compare.
edit: after some thought it turns out what I compared it to is literally floating point.
I would be interested in a real comparison between their floating-bar and true floating-point; it’s definitely not literally IEEE floating point, because their floating bar representation can represent rational numbers with an infinite representation in base-2, like 1/3. The fact that division of small numbers by small numbers has a perfectly precise result is a really nice property of floating-bar for some applications that floating-point can’t satisfy, but I’m sure floating-bar comes with its own downsides.
His final solution seems to be very similar in motivation to fixed point arithemetic, but with a “floating” point. I wonder how those approaches compare.
edit: after some thought it turns out what I compared it to is literally floating point.
I would be interested in a real comparison between their floating-bar and true floating-point; it’s definitely not literally IEEE floating point, because their floating bar representation can represent rational numbers with an infinite representation in base-2, like 1/3. The fact that division of small numbers by small numbers has a perfectly precise result is a really nice property of floating-bar for some applications that floating-point can’t satisfy, but I’m sure floating-bar comes with its own downsides.