/sci/ - Science & Math

>>15855169
Base 3 is actually more efficient, and rounding off expressions you just need to truncate the number, rounding in binary potentially requires the whole string of digits
https://earltcampbell.com/2014/12/29/the-setun-computer/
>Many basic arithmetical operations are particularly simple. Rounding of a number to leading significant figures can be achieved by just truncating a sequence, whereas binary rounding potentially depends on the whole sequence of bits. In The Art of Computing, Donald Knuth gives a fantastic survey of different number systems, and was so enamored with balanced ternary

>>15844944
Is fark or ats better? Or is there another?

>>15810042
https://github.com/openai/baselines/issues/805#issuecomment-469042844
>If you want exactness, then it's self-evident that computation with rationals is independent of base
Base 3 is actually more efficient, and rounding off expressions you just need to truncate the number, rounding in binary potentially requires the whole string of digits
https://earltcampbell.com/2014/12/29/the-setun-computer/
>Many basic arithmetical operations are particularly simple. Rounding of a number to leading significant figures can be achieved by just truncating a sequence, whereas binary rounding potentially depends on the whole sequence of bits. In The Art of Computing, Donald Knuth gives a fantastic survey of different number systems, and was so enamored with balanced ternary that he said it is “perhaps the prettiest number system”.

Another area is requiring determinism, cpus are deterministic, but gpus are not. If one needs to verify computation by repeating the same script, the outputs must be the same. The lack of appreciation of digit expansion has lead to many confusions in programming