Uljahn: to compare and select the highest?
Dlx: and why can't I get to finish my code when the clash is over, come on... I bet 95% of people here care most about solving it, not really winning it... so unsatisfying that i copied my code and finished it in a jupyter notebook lol
Dlx: Dang I'm doing this coding style for my next pair programming... will be working so fast and dirty on first pass that buddy will have no chanse to follow what I'm doing, just for fun, and then refactor refactor refactor after ;p
Uljahn: Automaton2000: fix it
Automaton2000: what if you could do a better job
MSmits: guys, I have an othello version that can take dbdr off nr 1
MSmits: it just wont be me up there, but tric trac :P
MSmits: 25% win vs you, 11% vs tric trac
dbdr: OK, what do you want MSmits? ;)
MSmits: nothing, i am not submitting this :P
MSmits: its just funny
dbdr: I thought you were over 50% against me
MSmits: but i can spam you down :P
dbdr: but I would just need to resub, no?
MSmits: its often like that
dbdr: check what trictrac offers you ;)
MSmits: i have no idea why btw, i just try to fit the params and these versions pop up
MSmits: it also does 50% vs jacek which is ok, but does poorly vs Robo
dbdr: probably hits a blind spot
dbdr: an opening bool would probably help with that
MSmits: win = true ?
MSmits: an opening bool
dbdr: book :D
MSmits: yeah that would probably help yes
MSmits: how much space do you have left?
MSmits: opening books dont require that much space if you're selective with what you put in
dbdr: I could probably compress source more if needed
MSmits: ahh ok
dbdr: and yes, a targeted book would be small
whay: i love axis
MSmits: btw this does assume your error happens in the first 20 plies or so
dbdr: given my lead, I'm happier to have a compiled version ppl can test against
dbdr: I've also looked at binary compression
dbdr: interesting rabbit hole
MSmits: though it's not unlikely. My bot is not very good compared to yours so it should be making mistakes fairly soon
whay: you have boot?
MSmits: you probably wont even get near your mistake when you make a book, because you already exploit an earlier one of mine
MSmits: yeah compression is interesting
MSmits: I was looking at the EDAX eval and figuring out if i could compress that
MSmits: but its just a tad bit too large I think
dbdr: I was talking about compressing the executable
MSmits: ohh ok
MSmits: well you have to be able to decompress and run any other precalculation in 2 seconds right?
dbdr: decompression is usually fast forCG-sizes, unless you go into NN type compression :)
MSmits: whats different about that?
dbdr: the (de)compressor is a NN
MSmits: ohh ok
dbdr: are you CG-vacationing? :)
MSmits: pretty much, i do have 2 things I have to do for work, but most of the time will be CG
whay: you can listen me
MSmits: I was on vacation last week
dbdr: I thought you were traveling
dbdr: ah ok
dbdr: that's why it took so long ;)
MSmits: my wife tells me not to advertise it online because of burglary
MSmits: i was actually working on othello during my vacation :P
MSmits: most importantly I made stuff into AVX
whay: you have wife
dbdr: it's true your real life collides with C because of your students, right?
MSmits: with C?
MSmits: hmm rarely yes
MSmits: might happen more often in the future
dbdr: collisions with C are dangerous, because it does not have much safety features ;)
MSmits: that's true I guess
MSmits: the students i get here are really not that type though
MSmits: CG is more used as an extracurricular activity
MSmits: since they're mostly not good enough to do stuff here
MSmits: it's funny, I had one student here a week or two ago that wasn't even in my CS class, I taught him physics
MSmits: he didn't pick CS because the previous curriculum seemed too easy for him
jacek: im still at figuring out what patterns would be useful
jacek: dbdr your 300k weights, those are NTuples?
MSmits: jacek why not look at edax
jacek: edax too complicated
MSmits: not really, there is 1 pattern you wont be able to fit
MSmits: you can fit 9-tuples, 10 is too much
MSmits: also edax has different eval for every ply
MSmits: just dont do that
MSmits: yeah jacek
MSmits: 3^9 bytes is doable
jacek: that would require fancy compression
MSmits: that's 10k characters
jacek: it has just one 3^9?
MSmits: one or two, not too many
jacek: is it 3x3 corner?
MSmits: think it might be yes
MSmits: eval.c is the file
MSmits: you can find out yourself
jacek: my preliminary results didnt show much gain
whay: msmits telle me
jacek: but perhaps i need to gib more training to it
MSmits: how fast is your bot with that eval?
MSmits: how many states does it eval?
jacek: 30k evals per turn
MSmits: ahh, so somewhere between Robo and mine
MSmits: kind of a halfway NN
jacek: 4 8x1 patterns, 16 lookups
jacek: + mobiility and frontiers
MSmits: Robo: 3k, you: 30k, me; 1 million
jacek: work smarter not faster
MSmits: mobility and frontiers I tought is supposed to be replaced by your eval?
MSmits: i think edax does just his ntuples
jacek: it improved the strength
jacek: perhaps those see game as global and patterns are more local
MSmits: might be because the ntuples just arent trained enough, or the patterns not complicated enough
MSmits: could be your approach is best for CG
jacek: the perks of machine learning. you dont know if there is bug in the code, or just training framework is bad, or you need to tweak the hyperparameters or change the model whatsoever
MSmits: yeah that seems annoying
jacek: and i recall tdgammon gained strength too when they added mroe advanced features to the raw board network
MSmits: your argument about global vs local makes a lot of sense
MSmits: mobility especially
MSmits: because all it does is give you more choices
MSmits: i dunno about frontiers though, havent been able to make it work very well.
MSmits: i am refittign mobility now, will try frontiers after, it's at a low value
jacek: it works for me, it always gives quite big negative weight
MSmits: hmm ok
MSmits: it's at -100 now for me. With mobility at +900
MSmits: probably i just havent explored it enough
MSmits: it's probably heavily correlated with mobility also
MSmits: might need to fit them together
jacek: its -0.057, while mobility is +0.11
MSmits: ok so half the size
MSmits: will try that later
jacek: ymmv of course
MSmits: all the versions I get lately are mostly strong against you and Renard, weak against the rest. Maybe telling me this helps you :P
jacek: and yes, those 9 are 3x3 corners
MSmits: ahh ok
MSmits: the smaller ones may be of interest?
MSmits: the release section also has a dat file with actual fitted values if you can decipher it
jacek: could be, so far i dont have any diagonals
MSmits: you might have to pick ply 20 or 30 or so
MSmits: because it is fitted by ply
MSmits: they use 16 bit values
MSmits: but the largest value is 2k something, so divide everything by 10.3 if i remember correctly
MSmits: assuming you use bytes
MSmits: I might have gone too far down this rabbit hole on my vacation :P
jacek: i fell in NN rabbit hole and i cant get out
YCS-Venom: me too :)
beyonik: why do my test cases keep expecting Nothing sometimes and other times they expect what I output
MSmits: low expectations?
MSmits: you obviously have a bug
beyonik: what bug
beyonik: how is it my fault
MSmits: and it can be many things
beyonik: that the test cases sometimes expect Nothing
beyonik: and other times they expect something
MSmits: i dont think they expected nothing, maybe your input/output is not synced?
MSmits: as in, you provide output when you havent gotten new input yet
MSmits: also, i dont know which puzzle or multi this is
Astrobytes: Or an extra newline, missing newline, one answer when multiple are expected, the list goes on...
MSmits: so its hard to comment
MSmits: what Astrobytes said
Astrobytes: But of course, it's the puzzle that has the bug. Clearly.
MSmits: thats probably it, something like that
beyonik: It surely wasn't newlines
beyonik: also they only expected one answer
beyonik: not multiple
MSmits: link the puzzle
beyonik: thing is I tried it for the first time, it didnt work, then I tried it again to see if I fixed my bug and it started just expecting "Nothing"
Astrobytes: Yes, it does that when your output is incorrect.
jacek: wooo another cat avatar
jacek: MSmits do you expand every children at once? as in, eval every children at once?
MSmits: I do yes
MSmits: That seems often better even when you do a full random rollout
MSmits: so it makes sense that it would be better for a quick eval also
jacek: so how do you count visits then
MSmits: visits += childCount
MSmits: well you get the point
jacek: then in that case i have rather 300k games, not 30k
MSmits: ahh ok, that makes more sense if your eval is fast
MSmits: to be more precise
tomatoes: its sometimes weird, but not difficult
jrke: can anyone make me explain movement of ant in langton's ant?
itzblinkzy: share code please
Illedan: jrke, I have no idea why you have such low time in SR :( But I heard others complained in other games too..
jrke: yeah dunno why i can use only 37-38ms
jrke: there is bug in csb also maybe i can use only 55-60ms in csb also :(
Illedan: I guess something CG did, added some weird overhead on the bots :shrug:
jacek: welcome scarfield
jrke: np i will manage something
Crypticsy: hey is anyone here really really good in bash?
jacek: is anyone ever good in bash?
jrke: btw how much time you are using in SR illedian
Scarfield: the level just increased by a lot
Crypticsy: so one guy tried solving my Van Eck's sequence puzzle and pointed out that he couldn't pass the final test
Crypticsy: i tried myself but couldn't either
Crypticsy: would you give it a try and tell me if it is possible?
Crypticsy: (in bash)
Illedan: jrke, what do you mean by time? Time to create it? Time to play it myself?
jrke: i mean like im using 37-38ms.
Illedan: Ah, I use 35 ms
Illedan: My code usually converges way faster anyway..
jacek: java? maybe too long warmup or gc
wlesavo: Crypticsy the last one requires optimized performance, last test works fine
Crypticsy: so you solved it in bash?
wlesavo: oh, sry, didnt notice bash
Crypticsy: XD also i am the one who created it
wlesavo: oh lol
wlesavo: i need to learn reading at some point
Crypticsy: yeah and i need to learn to explain bettter
itzblinkzy: SHARE YOUR CODE!!
ARatherLongUsername: prech itzblinkzy
Logiman09: What dis?
AbundantPuddle: Dis codingame.
_Akira_: how do you friend request someone?
dtinth: oops, the chat pops up while I am in CoC
dtinth: it moves the focus to the chat input, which is a bit unfortunate
AbundantPuddle: Yes, that has happened to me as well.
AbundantPuddle: I believe it is intentional to keep you on your toes.
whay: i have chanel
itzblinkzy: whay why you keep quitting clashes how will you learn
itzblinkzy: give it a try at least
whay: what do you mean
whay: level up
jacek: good for you
Astrobytes: good submit for robo in Oware
eulerscheZahl: top 2500 in CoC :trophy:
dbdr: top NaN in CoC
eulerscheZahl: you didn't play a single clash?
dbdr: should I?
eulerscheZahl: not much interesting content you are missing out
eulerscheZahl: except for the ascii value challenges
dbdr: it'll be needed for the quest map though
DomiKo: i wish there was no "shortest mode"
eulerscheZahl: i an live with a not completed quest map on my profile page
eulerscheZahl: an => can
Astrobytes: hacker.org had a quest map of sorts iirc
dbdr: it might block progress
dbdr: not a Power User without CoC ;)
dbdr: (that was a prototype, of course)
eulerscheZahl: i miss hacker.org :(
eulerscheZahl: countless hours of challenges
eulerscheZahl: i learned a lot of stuff there. like sending web requests or image manipulation
eulerscheZahl: that one was quite easy for me, once i finally unlocked it :D
eulerscheZahl: define start cell. then always flood fill to find the next bit. expand from the center of that region to find the nearest interesting color of another bit
eulerscheZahl: http://www.hacker.org/challenge/img/spiral3.png that was the crazy one :D
eulerscheZahl: after spiral bits there are some more fun ones. like a zip file with 4800 small images. you have to assemble them together to a large image
eulerscheZahl: just to unlock a harder version with about 18k pieces
eulerscheZahl: but these puzzles were always easier for me than the obscure crypto/stegano ones
eulerscheZahl: http://www.hacker.org/challenge/img/the-maths.png this one was cool too, having an extra twist after you solve the first part of it
DomiKo: 1. pick the game 2. do your best 3. read forum
jacek: csb can be done with few ifs up to gold
DomiKo: pick something you like
DomiKo: that would be the best choice
jacek: i.e. paper soccer :v
Astrobytes: Naturally :smirk:
AbundantPuddle: Clever TNtube
TNtube: Yes what ?
AbundantPuddle: Your answer on the clash. It was clever.
TNtube: What was the exercice ?
TNtube: I forgoted it xD
SirTarragon: Is there a way to grab the opponents distance to his checkpoint for csb?
JakeCubes: how does and AND work?
SirTarragon: && works as if (x1 == x2 AND y1 == y2) then when both x1 == x2 and y1 == y2 are true, then what's inside the if statement will run.
SirTarragon: Otherwise it won't.
SirTarragon: if(x1==x2 && y1 == y2)
jrke: good mningor
SirTarragon: Ugh was just doing the Power of Thor Episode 1, was trying to figure out what was wrong. Just read that X, Y starts in the top left instead of bottom left so had to change some calculations around wooo
SirTarragon: Mainly flip alot of my newYs from += to -= and the opposite
jokzyz: yo what is dn aio everybody keeps talking about it
el19oc: deez nuts