01:43 kylese left 01:44 kylese joined 02:07 vasko45355861 left 02:15 kylese left, kylese joined 05:33 Sgeo left 06:17 lizmat joined
disbot <melezhik.> Morning. If someone wish to check out dsci please let me know. Extremely fast and lightweight git server and cicd. Fast ci engine on Rakupp. Ability to port existing remote git repos . CI SDK supports many languages including Raku. CI jobs run on podman / docker or localhost . Authentication in git push over https 06:21
06:23 lizmat left
disbot <melezhik.> Current demo stand dsci.sparrowhub.io:8080/ runs on Rakupp , I need to tweak official documentation to reflect Rakupp migration changes ( not a lot though ) 06:24
<melezhik.> weekly: programming.dev/post/56332254 06:58
notable6 melezhik., Noted! (weekly)
07:27 lizmat joined 07:28 lizmat left 07:38 soverysour joined 08:10 dakkar joined 08:14 hulk joined 08:15 kylese left 08:48 eseyman left 08:54 manu_ joined 09:05 manu_ is now known as eseyman 09:09 soverysour left 09:39 soverysour joined 09:55 soverysour left 10:31 soverysour joined 10:44 soverysour left 11:03 soverysour joined 12:09 soverysour left 12:16 soverysour joined
[Coke] ok, that commit from august also just failed a blin run. :| 12:39
so I don't think it's one of the commits that's causing the trouble.
bah. 12:40
ww
12:56 soverysour left 13:09 soverysour joined 13:14 soverysour left 13:41 soverysour joined, soverysour left, soverysour joined 13:45 soverysour left
ab5tract TIL unums / posits 14:22
I can't help but wonder what using these in place of Rats would do for performance 14:23
[Coke] they're not actually rats, but "better floats"? 14:53
ah, they might be more Rat and less BigRat. 14:54
(sorry, skimming wikipedia right now :)
ab5tract yeah, I'm just digging in myself, it's a weirdly ingenius idea 15:16
the kind of idea where it's sort of like, why didn't we think of that before? 15:17
Of course, no general purpose hardware provides any support whatsoever 15:26
still, it could be interesting to at least benchmark one of the C libs. Depending on the threshold of where they fall stop being Rat-accurate, we could probably just let FatRats cover the distance 15:27
Voldenet posits are the future, but not near future 15:32
15:34 Sgeo joined
Voldenet though posits aren't replacement of rational or bigints at all 15:34
they're more like a replacement of `double` suitable for a lot of cases but also unsuitable for other (you still keep 8 bytes of info, you have to discard something to fit something else) 15:35
ab5tract Voldenet: if 0.1 - 0.1 == 0 then I don't know if that distinction matters
but as I mentioned, these are brand new to me 15:36
I guess what I'm trying to say is that we chose rationals because floating points were imprecise and they were the method at hand to achieve that 15:38
Voldenet in doubles you can represent things like 1 + 1e15, in posits you can't
ab5tract *that precision that we were seeking
Right, it's all about the 0^..^1 of it all. you make a great point 15:40
do you have any experience with them? it seems like not much has made it out of academia so far. 15:41
Voldenet I've experimented a bit with them, they're overall great for "real world data" 15:43
but for things where computations truly need dynamic ranges, it lacks dynamic precision 15:44
doubles are great for a lot of simulations where tiny fractions have impact on the whole system 15:46
like astronomy or physics 15:47
but posits are better for almost everything else :P 15:48
I remember that this lecture has nice plots explaining precision tradeoffs www.youtube.com/watch?v=aP0Y1uAA-2Y 15:57
(cw: awful dubbing) 15:58
ab5tract it seems counterintuitive to me that a variety of float could fit into a space where tiny fractions matter, but I'm essentially math illiterate next to many 16:01
Voldenet well, some good calculators handle scientific notation and it's good at explaining normal floats 16:07
e.g. you get 10 decimal digits and exponent that's -99 or +99 (no display for anything else) 16:08
posits have additional part that could convert one "9" out of the exponent into regular digit 16:12
or one digit into larger exponent
to me it's kinda intuitive if I use that calculator analogy (even if it's not directly how it works) 16:13
16:33 dakkar left 17:29 grayeul joined 18:29 ckfl3266 joined 18:31 dicit left 18:34 grayeul left 19:50 lizmat joined 20:01 lizmat left
ab5tract interesting, thank you for sharing 20:35
21:01 lizmat joined 21:37 Sgeo_ joined 21:39 Sgeo left 22:11 tellable6 left, kurahaupo left, tellable6 joined 22:19 kurahaupo joined