IOI2020: how to get more xp
Uljahn: solve puzzles, promote in multiplayers, unlock achievements
struct: morning
thanhhv317: here is afternoon
cnmdcbg: hey
TheBigBadGhost_936a: hello
cnmdcbg: 你好
jacek: good morning
AI_says_hi: o/
zhoubou: :hand_splayed:
ivnbcr: hi all
Passifi: top of the morning to you fellas and felicitases
Angecide: morning all, I have been trying to understand mcts solver lately, so my ultra high level understanding is that it is just a regular mcts until the a node in the tree reaches a terminal stage, in which case the algo tries to prove/disprove the move path? does this sound correct?
Angecide: since at the start, it is impossible to prove anything since the search space i huge, but once u reach mid/lategame, the search space is small and the algo can start to get more confident
Angecide: and I am still gold in uttt, turns out my huge performance boost didn't add anything since my mcts algo is not converging that well, I think it is too random in design + perhaps I shouldn't have used a xorshit8bit prng since it isn't really that random + potential bugs
struct: how many rollouts do you get?
Angecide: 40k or something, so something is definitely wrong
jacek: try xorshift64
jacek: and those rollouts should be enough for legend
jacek: as for mcts solver. what do you do if you encounter final state in selection?
Angecide: I just return the value back up, as if it was the result of a simulation
Angecide: like for my normal mcts
jacek: ok
Angecide: I am still in the process of understand what mcts solver is and how it differs from the normal algo
jacek: my solver gave 55-60% winrate, but still it should be enough for legend
Angecide: ye going through all the replays my ai played, I found one instance where I did an "invalid action", so I need to go on a bug hunt, since I presume my algo is converging badly because of false premise of the state of the game
jacek: happy bug hunting then
jacek: mcts solver is quite simple. it is just the mcts solver is complicated
jacek: mcts olver paper*
k4ng0u: Angecide, my bot only have around 8k per turn and passed to Legend. Do you do full random rollouts or do you have some early exits? typically in my rollouts when I have a winning move, I ignore the othres and just play it. This prunes a lot of unrelevant states
Angecide: I do a very bad (but fast) pseudo random rollout, but I believe adding heuristics such as always playing winning moves and avoiding losing moves could help too
Uljahn: also exploration coefficient can cause bad convergence, have you tried to vary it?
Angecide: ye, so my current definition of evaluating whether or not my mcts is converging is by checking if it forces the enemy player onto the same board (if possible) during the very early stages of the game
k4ng0u: avoiding losing moves is a depth 1 simulation for each move, it would be too time consuming. Also note that it would be covered on your opponent turn by always playing winning moves
Angecide: like even after 40 secs of searching, it doesn't recommend forcing enemly player onto the same board, so i suspect something is wrong
codybumba: If anybody is interested in CoC please join this private clash - https://www.codingame.com/clashofcode/clash/1544652d4ca36a11f6da00d0a8255e74cdeb9db
darkhorse64: This is an heuristic which took years to find so it is not surpising your bot does not find it. Re solver, remember you can backpropagate solved nodes: if you have a win, it means the parent is a loss, if all children are losses, it means the parent is a win and so on. This way you can predict a game result 10 to 20 plies in advance
jacek: darkhorse64 my bot can find it most of the time with less than 5 seconds
Angecide: ahh I think it is making sense now, it was a bit unclear from the paper, but I am starting to grasp what mcts solver is and that it is really not that complicated
jacek: another mcts solver paper victim
darkhorse64: Only the first 100 ms matter to me:grinning:
darkhorse64: Basically, that's what the paper says but it does not really need several pages
darkhorse64: low cost optim, high reward
Angecide: ye now with the intuition in place, it makes sense when reading the paper again
darkhorse64: jacek: you seem to have improved your C4 bot lately. Did my code help ? Did you see any difference between AVX and scalar ?
jacek: i trained n-tuple. i switched from pure mcts to jacekmax
Angecide: jacekmax :thinking:
darkhorse64: Great but that does not help me. I have some planned improvements but there is no point to do it now before approval. How many weights ? Straight or random ?
Angecide: nvm I found the playground article
jacek: every row-of-4, theres 126 of them. i didnt reduce for symmetry
jacek: and i didnt try your code
darkhorse64: To make things clear, I would have had no problem with it. I shared it after all
jacek: well if you would have problem, i wouldnt confess :)
darkhorse64: :relaxed:
jacek: meh i registered to the computer draughts forum, but it is admin activated :c
Kokoz: anyone want to help with a noob question?
jacek: you mean 2 questions
AI_says_hi: lol
Kokoz: no, only one
AI_says_hi: kokoz u already asked one ;-)
darkhorse64: ^
darkhorse64: kokoz: go ahead
jacek: meaybe he really meant 1 question
AI_says_hi: hehe jokes on us
Kokoz: so, I have a list [2, 3, 6, 7]
How to get the diffference between every number and all numbers before it??
like this:
3-2, 6-3, 6-2, 7-6, 7-3, 7-2
AI_says_hi: Think about using the modulo operator
AI_says_hi: Actually no I'm dumb :) think about using nested loops
darkhorse64: is it for the horses race ?
Kokoz: I am actually trying to make a nested loop but I dont know how to write it :\
Kokoz: No.. I finished the horse race puzzle
darkhorse64: a for loop inside the for loop
darkhorse64: for() { for () {} }
struct: you make a inner loop that goes from current pos to position 0
struct: i=0;i<end;++i
j=i-1;j>=0;--j
Kokoz: I tried something like that but I guess I wrote it wrong.. I will play around more, thank you
AI_says_hi: Kokoz what language are you using?
Kokoz: Python
SPDene: http://chat.codingame.com/pastebin/651ccb83-2ab5-4298-b013-d410619e9bd2
SPDene: kokoz ^ that was message for you
AI_says_hi: How do you make such pastebins?
SPDene: just paste enough text, and it will auto-convert it
AI_says_hi: oh okay thanks!
Kokoz: Ohhh, thank you SPDene, I was so close with my code :D
My mistake was that I put the a = x[i] in the second loop.. Which was dumb
Kokoz: Soo, everything is fine untill the input has a huge amount of information.. Then my code cant handle it
struct: yeah, you can do this with 1 loop only
struct: oh wait
AI_says_hi: What puzzle are you on?
struct: maybe not
Kokoz: Stock exchange losses
SPDene: not seem that one before - but I don't think you need the differences for ALL the previous values - just the *highest* value efore now
AI_says_hi: ^ thats right
Kokoz: so in that case [1, 2, 4, 4, 5]
what will be the result?
AI_says_hi: 0
AI_says_hi: Are you sorting the values?
Kokoz: no
AI_says_hi: In [1, 2, 4, 4, 5] there is no drop, all numbers are >= to the one before
AI_says_hi: so highest loss is 0
Kokoz: Exactly
AI_says_hi: So then what is the problem?
Kokoz: I still have to use loop to check all other numbers
Kokoz: I have an idea
Kokoz: so if the input is sorted then in all cases there is no lost..
Kokoz: Nice, thank youu C:
AI_says_hi: did you manage? :)
Kokoz: not yet, but on the right track :D
AI_says_hi: good good
Salman2301: https://www.codingame.com/clashofcode/clash/15448669b288f34396c1947508d83db43ea4c6b
Salman2301: join
Murleys: Why can't we choose which mode to play in clash of code ... I've had shortest code 5 games in a row and its old ...
jacek: only in private clash
Salman2301: https://www.codingame.com/clashofcode/clash/1544885a66ac2c22cd8554757e1287b6a7a326d
Salman2301: I had the fastest mode I am just not ready for that
lopidav: >Why can't we choose which mode.
I guess the whole system will have to be redesigned. People would learn Ruby/Perl and play shortest exclusively
RoboStac: plus there are already bots to get enough players for it to work without splitting it further
READY: I am READY for that, Salman2301
TheMightyTopHat_dc86: Me i am all out
Pad0000001: In the AI pod racing stuff, can i check out others' code?
jacek: CSB? no
JM54: Soon may the wellerman come to bring us sugar and tea and rum
Astrobytes: WTF. Since when do people sing sea shanties on the reg.
JM54: Since today
jacek: mhm https://i1.kwejk.pl/k/obrazki/2021/01/PDh9PlsKF97hY8ZZ.jpg
Astrobytes: lol, that's an old one
jacek: youre cheating on me with another meme source?
Astrobytes: No I remember the story, from years ago. Something about bood types iirc
Astrobytes: *blood
eulerscheZahl: moin
Almost404: hello guys im new here and i am having tones of fun
Almost404: this is great
Astrobytes: afternoon euler
Almost404: sad i didnt see this sooner
Astrobytes: Well it's not gonna disappear so you still have time Almost404
eulerscheZahl: you mean: "sad I didn't see this soccer"
here it is: https://www.codingame.com/multiplayer/bot-programming/paper-soccer
Almost404: im happy i also made today my first problem contribution anyone wants to see it?
Astrobytes: oO
JM54: As a beginner in code and barely starting to learn python, the puzzles in this game are way beyond my understanding
Astrobytes: JM54: the site does assume familiarity and a certain level of competence in your language
Astrobytes: It's not an absolute beginner-friendly site.
JM54: dang
JM54: sucks i got assigned this then
eulerscheZahl: who assigned it to you?
eulerscheZahl: a teacher?
JM54: My CSP teacher
bobth: some puzzles are super easy but most will need some experience imo
Astrobytes: Just practice with the language a bit more and you should be able to do some of the easy puzzles
JM54: Noted, thanks
SPDene: also be aware that some "easy" puzzles are a lot more difficult than others
Astrobytes: ^
JM54: Ok
darkhorse64: However, a good advice is to separate coding from algorithm. Do some paper reasoning first and then, only then, translate it into code
bobth: pseudocode is very helpful
JM54: :thumbsup_tone3:
BlaiseEbuth: Paper ? What is this ?
eulerscheZahl: or less fun. even thibaud mentioning Defibrilators here
Astrobytes: Yep, very sound advice.
Astrobytes: lol euler, right
JM54: onm
eulerscheZahl: i'd add aneo to that list
jacek: hm?
eulerscheZahl: BlaiseEbuth I found that paper: https://sketch.io/sketchpad/
darkhorse64: There is no data parsing which is really the worst
jacek: a sketch? like that one before?
AnalPainGiver: https://www.codingame.com/clashofcode/clash/15450015d77040b815e7a2b1ddd9437ec95922e
Astrobytes: you still have that jacek?
BlaiseEbuth: Thanks eulerscheZahl, I was wrong, I thought that paper was slices of innocent tree corpses...
jacek: on my personal comp, not here
eulerscheZahl: i will neither complain about the clash link nor the username
jacek: Oo
eulerscheZahl: we want to engage more newcomers :)
Astrobytes: Same.
BlaiseEbuth: Frfllf... :shrug:
jacek: i see smits still posting c4 moves
jacek: and 0 seems to be losing after all
Astrobytes: Is that the move that was oscillating?
jacek: yes
AnalPainGiver: I am looking for enaging clash of clans where all the opponents are live on twitch
AnalPainGiver: code*
jacek: oh my
BlaiseEbuth: your what ?
Almost404: is this too hard for a clash? https://www.codingame.com/contribute/view/609613f0fb1078f73696777d6b128c5ae046
ChampionCoder: I felt it was, don't strike me down for that
ChampionCoder: I felt it was hard
Almost404: check it again I redid the solution, literally the solution is 1 line after data ready
ChampionCoder: Oh ok
ChampionCoder: :+1:
ChampionCoder: looks better now
eulerscheZahl: "Every quarter of what's left he stops for air because he is fat."
isn't that an infinite sequence?
Almost404: tks maybe i shouldn't have made a detailed solution it made it look it was harder than it actually was
Almost404: that's true let me fix that
Uljahn: kinda hard yes, 'cause you need several minutes just to read and understand the statement
SinOfWrath: Also by doing integer division in your solution, I'm fairly certain it's incorrect when speed is not a divisor of 100.
SPDene: Almost404 "Velocity = Time = Distance/Velocity" confuses me. did you redefine the laws of physics? :P
ChampionCoder: :sweat_smile:
Almost404: sleep deprivation is getting me :joy:
ChampionCoder: By the way, why is "fat" in bold?
Just make it normal, not bold
Almost404: discard this shiet i tryed need to review this
eulerscheZahl: "fat" in fat text? makes sense to me :P
ChampionCoder: Lol
kovi: old john...an impressive start on cg (and nice background, started programming/boardgames earlier than me)
ChampionCoder: Sorry?
Astrobytes: He worked alongside Buro and Le Renard
eulerscheZahl: Buro?
Astrobytes: Michael Buro
darkhorse64: We are getting more and more smart people
eulerscheZahl: i guess i'm not up to date with the users anymore
Astrobytes: (of Logistello fame amongs many other things)
Astrobytes: Did Le Renard create Thor?
Astrobytes: (othello program)
jacek: this old john was having books wars with smits in othello
Astrobytes: Still is afaik
kovi: i have just noticed him, as he jumped to #3 in d&b
jacek: still losing, yes
MSmits: he stopped for now. I think we agreed to a draw
MSmits: (not sure though)
eulerscheZahl: a draw with you above him on the leaderboard
MSmits: i do better against other players
MSmits: interesting about D&B
MSmits: I'll try a resubmit with random on. My current submit is deterministic
MSmits: (this version did better vs remi)
MSmits: i should say, i use fixed seed currently, i'll change to a time based seed
eulerscheZahl: so the survey results are out
eulerscheZahl: https://www.codingame.com/work/codingame-developer-survey-2021/?utm_source=community&utm_medium=cg_chat&utm_campaign=2021-developer-survey-report&utm_content=main-batch
eulerscheZahl: i modified the URL a bit to confuse their tracking of how we got there :P
MSmits: is codingame the nr 1 tech hiring platform?
MSmits: or is this just advertising?
eulerscheZahl: yes. source: CodinGame
MSmits: lol
eulerscheZahl: there are some other big players like Hackerrank and Hackerearth
MSmits: I see
MSmits: isnt linkedin a tech hiring platform also, technically?
eulerscheZahl: not sure if CG really is the #1. by what metric even?
eulerscheZahl: hm, good point
eulerscheZahl: maybe tech hiring = coding assessments in the understanding of CG marketing
MSmits: could be
eulerscheZahl: let's just say: be careful what you read and don't believe anything
MSmits: oh I am that, I was just wondering
eulerscheZahl: so: don't believe me when I say that
MSmits: surely you are an exception
eulerscheZahl: the truthful toad
MSmits: yeah
eulerscheZahl: now i'm still at page 1 of that survey
MSmits: sorry
eulerscheZahl: "48% of companies offer developers the possibility to work 100% remotely"
how to measure that?
MSmits: maybe this just was a question?
eulerscheZahl: they only ask for the company size in that survey, not the company itself. how to find duplicate companies?
MSmits: oh
jacek: :thinking:
eulerscheZahl: because it's about companies that allow remote and not devs working remotely
MSmits: that might be problematic yeah
jacek: more remote last year i presume
eulerscheZahl: let's take notes for when Thibaud posts it on the forum or discord
MSmits: do you want to open a thread: "questions from the 5%" ?
eulerscheZahl: 5% that cause 50% of Thibauds work :P
MSmits: no doubt about that
darkhorse64: 50% headache also
jrke: can anyone just tell me that what happened yesterday brainstorming on discord in breif?
eulerscheZahl: too bad inoryy wasn't there yesterday :popcorn:
eulerscheZahl: Thibaud summarized it on the forum
eulerscheZahl: https://www.codingame.com/forum/t/community-regular-events-lets-discuss-it-on-discord/188710/11
MSmits: allright, so basically D&B is like: Miklla 1st, then the next 3 as a group. Therés no clear 2-3-4 order
darkhorse64: There have been some clarifications from Thibaud and G-Rom. One thing is sure; community contests are gone. There might more regular, fast events
eulerscheZahl: and you allow miklla to do that?
MSmits: eulerscheZahl not in the long term ofc :)
MSmits: D&B is a very complicated multi, i cant improve my bot ina day
eulerscheZahl: but these "fast events" don't sound like original content at all
jrke: MSmits thinks deep in life as well as in bot
eulerscheZahl: just some old meal in the microwave
eulerscheZahl: so at least I am not interested
jacek: fast event? clash wars?
eulerscheZahl: also
darkhorse64: No news from AshKetchum. No C4 yet
eulerscheZahl: some extended version of the puzzle of the week with more featuring
MSmits: jrke thanks, i take that as a compliment :P
jacek: no news for the chess either
jrke: MSmits that was the compliment ,nothing else
eulerscheZahl: fireworks at 2 approval votes
eulerscheZahl: i just opened it to dive into it
MSmits: it's original enough
eulerscheZahl: then survey, then chat
eulerscheZahl: i'm easy to distract
MSmits: me too, I actually should go grade some christmas postcards
eulerscheZahl: you teach arts now?
eulerscheZahl: in elementary school
MSmits: mmh i guess kinda
MSmits: this is a python turtle assignment
eulerscheZahl: aah
MSmits: they had to draw a sun, trees, road, snow, text all with python code
eulerscheZahl: my first thought: there must be a 2nd meaning for "grade" but let's nag smits with it
MSmits: hehe it's just 3 possible grades
MSmits: fail, sufficient, good
MSmits: so it's not hard to grade
eulerscheZahl: do you also grade the code style or just the outcome?
MSmits: this is the very first time they code anything in their life basically
MSmits: so style is not really an issue
eulerscheZahl: except for indents
MSmits: well it has to work :P
eulerscheZahl: i'm on page 3 of the survey already
jrke: survey? which survey?
eulerscheZahl: https://www.codingame.com/work/codingame-developer-survey-2021/?utm_source=community&utm_medium=cg_chat&utm_campaign=2021-developer-survey-report&utm_content=main-batch
eulerscheZahl: i invented a new utm_medium key just to confuse Thibaud or whoever reviews that
MSmits: this is my D&B submit, pretty clear:
jacek: anyone expert with keras (python)? how could i efficiently train network with sparse inputs, like 64 ones and 1500 zeros
MSmits: http://cgstats.magusgeek.com/app/multi-dots-and-boxes/msmits
MSmits: wrecked by miklla
eulerscheZahl: when you have to clarify that keras = python, that user clearly isn't an expert in it
jacek: MSmits you get pwnd by miklla
MSmits: yes
jacek: dunno, maybe there is keras c#, i may never know
jrke: why my endgame solver is looking like this - https://www.codingame.com/share-replay/521887983
jrke: :(
jacek: interesting http://cgstats.magusgeek.com/app/multi-dots-and-boxes/miklla
MSmits: well you won that one though jrke
MSmits: jacek i actually know why he wins
eulerscheZahl: i'm a kera expert: kera finished 4th in the Ghost in the Cell contest
jrke: just random seed but my endgame solver is bugged
MSmits: it's very well documented how to play this game and it's hard to code a bot for it
MSmits: he counts chains
MSmits: you see he has a typical strategy of forming the right number of chains
MSmits: he doesnt just randomly place lines
MSmits: my bot is random for 50 plies
MSmits: of course that's gonna fail against anyone who knows what they're doing in ply 1-50
karliso: wow, clean 100% submits...
jacek: fix your uttt bot
karliso: :D
MSmits: oh, right karliso, i wonder if re curse 's bot is really stronger than yours or he just picked one that is strong against your current choice of bot
MSmits: try to make a small alteration
MSmits: then test it
eulerscheZahl: oh, he submitted his NN now
jacek: oh and im not the only one with misleading eval https://www.codingame.com/share-replay/521616666
MSmits: after all i was able to get 100% as p1 vs your current bot, so a trained nn may get lucky on one iteration also
karliso: yea, I looked at it. I think I had like 33% as white and 34% as balck...
MSmits: thats very interesting
jacek: see how robo in 54 says 0.92 while clealy losing
jacek: white black in uttt? o.O
MSmits: i am just guesing white is p1
MSmits: guessing
karliso: yeah, I tend to forget who is ) and who is X
MSmits: X is first :P
MSmits: it's weird though karliso, something is off here
MSmits: no way should your p1 bot have 33%
MSmits: not even vs a perfect playing bot
MSmits: the disadvantage of p2 is too large
MSmits: I am guessing almost every game is the same
MSmits: with 2 deterministic bots
MSmits: and you just make the same mistake every time
MSmits: or at least, maybe there are only 8 variations of game or something like that. Seen it a lot
karliso: anyway, I will have a reason to improve my bot
MSmits: yeah, me too, but still got other things to do first
MSmits: I want to fix my cotr and make a real solver
jacek: cotr?
MSmits: that bilbo optim
jacek: bilbo optim?
MSmits: make words
eulerscheZahl: coders of the realm?
MSmits: code of the ring?
Astrobytes: Code of the rings
eulerscheZahl: i blame Bob for reassigning that acronym
MSmits: my current submit is done by hand
MSmits: rank 52 with typed up solution :P
eulerscheZahl: mine is partially by hand
MSmits: i am guesing the loopy ones
MSmits: guessing
eulerscheZahl: sure
eulerscheZahl: didn't do the last testcase by hand :D
MSmits: I did :P
eulerscheZahl: what score on that alone?
MSmits: only took 3 hours or so
MSmits: not sure, but it's pretty good
MSmits: i thought a lot about reusing boxes and such
MSmits: pretty sure a simple beamsearch would beat it though
eulerscheZahl: you just have to run your code once to check :rolling_eyes:
MSmits: code?
eulerscheZahl: hardcode
eulerscheZahl: just execute whatever mess you caused
Astrobytes: lol
MSmits: lemme check
eulerscheZahl: and give me a number so i know if it's worth to improve that alone
eulerscheZahl: i'm at 1393
MSmits: ok sec
MSmits: 1402 frames, is that the score?
eulerscheZahl: no
eulerscheZahl: hover over that green testcase play thing
darkhorse64: I can see how a BS could minimize the steps needed to build the strings but you will need a post process for loops
MSmits: dont think the big testcase uses a loop anywhere
eulerscheZahl: yeah, those loops make it hard
MSmits: says the magic phrase is incomplete
eulerscheZahl: testcase vs validator
darkhorse64: Right.
eulerscheZahl: you didn't hardcode the test it seems
MSmits: no just the validators
MSmits: but the last one is identical almost
MSmits: rigt?
MSmits: right?
eulerscheZahl: no
kovi: luring of the dark side...
eulerscheZahl: last letter is removed in the validator
MSmits: yeah so almost
MSmits: how do you calculate score from character count?
MSmits: is it just the same?
eulerscheZahl: oh wait, that 1393 was an online solution
eulerscheZahl: because i didn't find it in my book
MSmits: the number of frames = character count
MSmits: so 1402
MSmits: i pasted in a text file
darkhorse64: right on the fact that there are no loops on the last sentence. Maybe for the ring word ?
eulerscheZahl: 1332 on the validator
MSmits: so my score is 1402 by hand
MSmits: pretty nice right
eulerscheZahl: my code is slightly better
MSmits: yeah
eulerscheZahl: also without loops
MSmits: i just did this because I wanted to get past the quest
eulerscheZahl: so i'm missing points somewhere else
MSmits: but then dbdr said, why not just do a beamsearch
MSmits: and i was like... damn
eulerscheZahl: but where is dbdr?
MSmits: dunno
Astrobytes: Still no word from him?
eulerscheZahl: i miss you, my little prince
MSmits: thats not creepy at all
Astrobytes: :o
eulerscheZahl: still running through dungeons
Astrobytes: lol
MSmits: is he playing Dnd?
Astrobytes: No...
Astrobytes: His avatar
eulerscheZahl: :facepalm:
MSmits: o right
MSmits: went woosh
eulerscheZahl: he's gone for so long, you forgot about him
MSmits: no no, just the pic
Astrobytes: Hope he's OK
MSmits: hes not been gone *that* long
MSmits: less than a week isnt it?
Astrobytes: No
eulerscheZahl: much longer
MSmits: hmm
MSmits: maybe it seems shorter because i havent been that active
MSmits: has he never been away before?
eulerscheZahl: not for long at least
MSmits: hm ok
MSmits: well i too hope he is ok, he's part of the CG furniture
MSmits: he's the chair, you're the couch euler
eulerscheZahl: and Automaton2000 is the old couch that no one wants to pick up even you placed it online as "free for pickup"
Automaton2000: do you guys know how to turn it into a pastebin
MSmits: he's the bin
eulerscheZahl: i understood the CG survey now. they also asked companies/HR guys. not only coders
MSmits: ah
jacek: did they hire AutomatonNN
AutomatonNN: cool
MSmits: I think they only hire non-binary
Astrobytes: He's here all week folks...
MSmits: :)
jacek: hm?
Sentinel25B: yo
Sentinel25B: whats up
MSmits: enum { up = 0, .. }
MSmits: hi :)
eulerscheZahl: int[] dx = {0,1,0,-1};
int[] dy = {1,0,-1,0};
MSmits: works
eulerscheZahl: https://yadi.sk/i/KjYmchxEreacg
MSmits: I imagine if my wife didnt order all my clothes I would only be wearing stuff like that
Astrobytes: lol, I forgot about your 'arrangement'
MSmits: hehe yeah
eulerscheZahl: you aren't allowed to look like a nerd? :D
MSmits: if I pressed the issue I suppose i probably could have one nerdy day a month, if i didnt have to leave the house that day
eulerscheZahl: https://www.getdigital.de/shop/kleidung
for your monthly needs
eulerscheZahl: but beware of the dryer
MSmits: good stuff
Astrobytes: You need the purple tentacle one MSmits
therealbeef: tentacle looks good for winter
MSmits: lol
Astrobytes: The Gandalf asymptote...
eulerscheZahl: https://www.getdigital.de/Debugging.html
i like those nonsense sentences
Astrobytes: Facehugger scarf, now we're talking.
Astrobytes: My favourite t-shirt currently is my "You read my t-shirt. That's enough social interaction for one day"
jacek: so nerdy
Astrobytes: And I didn't know it would be so apt for COVID-times when I got it.
miklla: someone summoned me by red signals :)
jacek: or maybe you knew it ?thinking:
miklla: found out D&B at end of december, love from first sight :)
miklla: as TRON :)
jacek: tron and d&b? huh?
miklla: multis
jacek: what do they have in common
eulerscheZahl: suum cuique
miklla: grids and simplicity of mechanics
jacek: 'simplicity'
Astrobytes: you and your latin euler
eulerscheZahl: sorry, was tortured for too long
jacek: quidquid Latine dictum sit, altum videtur
eulerscheZahl: also: Ich und mein deutsch :P
jacek: schmeterrling!
Astrobytes: charliebus sittibus on the deskinorum
miklla: spend around 4 days for huge improvement, commited, got 100% winrate as before, nothing changed :(
eulerscheZahl: "whatever is said in Latin will be seen as deep"
jacek: miklla did you read papers on d&b?
miklla: yes
Astrobytes: Is MSmits actually missing question time with miklla?
jacek: lets call him. "2 3 is good opening move"
Astrobytes: "meta mcts"
Astrobytes: "bitboarrrrdsssss"
reCurse: booking as p1 in uttt is viable
jacek: eris pads her chest
Astrobytes: He must truly be afk.
jacek: reCurse but it is. as p2 it isnt
reCurse: I see, you're trolling to wake him up as well
jacek: are you gonna release chess soonish?
reCurse: Need to write a boss for it
reCurse: And I'm distracted with bt
eulerscheZahl: you have random() already
jacek: another player for bt :scream:
reCurse: I've started training, quite a bit more resource intensive than uttt
reCurse: Need that new cpu and gpu
jacek: me too :(
reCurse: Either that or I make it cloud compatible
jacek: i wanted to replace my PC before the new ryzen, but no, "there will be new ryzen, price will drop and ill buy it then"
reCurse: It's a good thing you waited, zen3 finally fixed bmi
reCurse: 1000x speed increase or something
jacek: but now i still have no bmi
reCurse: On par with intel now
reCurse: Ah
jacek: but right now im gonna try those python tools for learning
jacek: i have data and i train using my half-assed made code
reCurse: Puck Fython
jacek: but i need a way to efficiently compute sparse input for those :c
reCurse: There seems to be a distinct advantage for p1 in bt so far
reCurse: It's a shame, there's no pairing
jacek: 5x5 and 6x5 is win for p2
reCurse: I don't care much for fun sizes though
reCurse: 10x increase in disk size to store training games compared to uttt lol
jacek: is your bot alphazero style?
reCurse: Kind of
jacek: im too dumb for training policy
Astrobytes: You always underestimate yourself jacek
jacek: dunning kruger eh?
eulerscheZahl: isn't that the opposite?
reCurse: It's both
eulerscheZahl: oh
MSmits: I'm here!
MSmits: miklla are you counting chains?
reCurse: Ping to Netherlands is crazy
Astrobytes: lol
MSmits: or doing some form of nimstring analysis
miklla: MSmits no
miklla: actually my around 39 first moves have no meaning
MSmits: i notice you have a particular way of placing lines early game
MSmits: ah yes
MSmits: i noticed from your output that you're a few turns ahead o fme
MSmits: sometimes it says loss
MSmits: and then turns to win
miklla: just not implemented euristics, make non dumb move with lowest index
MSmits: so something happened for your bot to think i won
MSmits: ah ok
MSmits: yeah i see the lines build up from the bottom up
MSmits: lowest index checks out
eulerscheZahl: so you were over-interpreting bot messages?
MSmits: no, the loss thing i sreal
MSmits: this happened around ply 45 or so
reCurse: I still need to get around with trolling my bot messages
miklla: these message come after ~39th move
MSmits: and then my bot only started solving after 50
miklla: only after*
miklla: so they have meaning
MSmits: yeah
MSmits: when your bot says "loss"
MSmits: is this a certainty?
MSmits: on perfect play?
miklla: yes
MSmits: interesting
MSmits: quite early
miklla: initially in loss position my bot was making random or lowest index move
miklla: but now it makes kinda best losing move
MSmits: always better to do that
MSmits: so is this like a negamax?
miklla: usually it can be countered by 1-3 of 30 available moves
miklla: and it gave me nice winrate boost :)
MSmits: or more like a statistical search with a solver
MSmits: (like mcts solver)
MSmits: i've been thinking of doing something with that, and improving my solving capabilities.
MSmits: currently i just do nimstring analysis until solved game, which basically means I start doing wel from 50-55, 38 is obviously much better
miklla: right now I have just endgame solver, nothing random or impartial
MSmits: does the endgame solver do score, or just win/loss?
miklla: just endgame starts around 39-43rd move for me :)
miklla: win/loss
miklla: in different game
MSmits: ah, that could make things faster for me, if i didnt bother to do score
jacek: i dont bother either :v
MSmits: it just looks cool to output the predicted score :)
miklla: remi sometimes win ,e because I win in other game, but lose in real game
MSmits: which other game?
miklla: well known game in D&B literature
MSmits: strings and coins? Thats basically the same game, but represented differently
MSmits: or do you mean something else?
miklla: nimstring if you want precise name
MSmits: ohh i got it
MSmits: yes sometimes you can win the nimstring and still lose on score
MSmits: so your bot only searches nimstring?
MSmits: so your endgame solver is a nimstring solver
MSmits: meaning you do nimstring analysis
miklla: y
MSmits: good yes, this is how i beat remi, i started doing nimstring analysis
MSmits: but not good enough apparently :)
MSmits: it's on my list to improve this. But I will work on other things first. Thanks for sharing miklla, much appreciated
miklla: ye, kinda boring with 100% winrate, have 7 ideas to improve my bot, waiting for losses :)
MSmits: do what i did, make a score solver. Once your bot solves the nim-game, you got calculation time left over
MSmits: then you can output score :)
MSmits: will matter almost never, except sometimes vs remi
miklla: remi is cool when he produces 2 cycles
MSmits: yes thats how you beat someone who won the nimgame, as many loops as possible
MSmits: but remi also gives away a lot of boxes, so he's usually already behind in score
MSmits: i wonder if anyone else understands this :P
eulerscheZahl: i gave up long ago when you are chatting
MSmits: :grin:
eulerscheZahl: so i just share this post that i found while searching for the dx dy tshirt: https://codeforces.com/blog/entry/62094
MSmits: this is when the police finally got to you euler?
eulerscheZahl: that's not how i looked
dkim19375: :grinning:]
MSmits: oh ok :P
eulerscheZahl: i think i shared a pi of myself a while ago, didn't i?
MSmits: you may have
eulerscheZahl: but you didn't see or already forgot again
MSmits: i know this was not you, but I dont remember what someone looks like from seeing 1 pic 1 time
eulerscheZahl: me neither
eulerscheZahl: i'm terrible with faces :(
MSmits: i have 100+ new student faces to remember also, I think the information gets drowned out
eulerscheZahl: one reason why i prefer a series over movies
MSmits: I'm reasonably good with faces, but not with the names attached to them
MSmits: so i will recognize a student years later and then not know their name
andrecab: Hi there! in the codingame C++ ide I have two classes that need to know about each other. My problem is that the class declared at the top will not know about the one in the bottom
eulerscheZahl: once I was like "why is this guy coming towards me" (just thinking, not saying). friend: hey, there's your dad. oops :D
MSmits: oh, thats a different thing though, that's attention
andrecab: normally I would use .h files but how can I solve this in CG?
jacek: andrecab isnt that true for any c++ ide?
MSmits: I've nearly walked by my wife and daughter several times on the street, just thinking and not registering their faces
miklla: you can forward declare classes and use pointers
reCurse: andrecab There is nothing special about #include files, they are literally pasted in the source file by the preprocessor.
reCurse: So whatever they did you can do the same manually.
MSmits: do conflicts happen often with includes?
JBM: a bit of magic for __FILE__ and __LINE__ hough ;)
MSmits: i had it happen a few times with string
reCurse: Define conflict
MSmits: same name for things in different include files
jacek: #ifndef
MSmits: then gets pasted in
reCurse: Only happens if you did it wrong
MSmits: no doubt
JBM: it's not really #include-specific either too
JBM: you'll get naming conflicts across translation units too
MSmits: right yeah, it's more general I guess
andrecab: Ok, thanks for the help!
MSmits: I had it happen with C# as well. I was using unity and tried using some basic C# libraries, apparently unity also had definitions for the same things and it caused conflict
reCurse: And that's how namespaces were born
reCurse: Or packages for java heretics
MSmits: right
JBM: or mangling
AnalPainGiver: Why my xp's are not increasing ? despite solving problem?
eulerscheZahl: is the problem a clash?
jacek: i feel butthurt just reading your name
eulerscheZahl: let's engage the community :)
reCurse: The name gave it away
eulerscheZahl: that's the 95% we want
JBM: i'm in bed and eZ's still active
JBM: something's messed up
struct: The icing on the cake will be the new feature
MSmits: AnalPainGiver, I think you had trouble translating from Indian to English
MSmits: it's called a hemorrhoid
JBM: like you typed that correct on the first try
AnalPainGiver: ?
eulerscheZahl: you mean the twitter post feature struct?
struct: yes euler
MSmits: i google it JBM
reCurse: He has experience
AnalPainGiver: Dude, MSmits
JBM: wait is that a real thing
AnalPainGiver: what are you talking about ?
MSmits: it's your name
JBM: what ELSE did I miss from that infamous chat
AnalPainGiver: I just asked a generic question.
MSmits: no no
eulerscheZahl: and some new tile at the top announced in a few days or up to 2 weeks. wonder if those are about the same thing
MSmits: i was responding to your name
AnalPainGiver: Oh ok, I love doing anal to girl, that's normal!
AnalPainGiver: If you don't, that' s on you.
JBM: as long as she loves it too
AnalPainGiver: True, JBM
MSmits: uh oh, I'm out
jacek: oO
eulerscheZahl: the 95% of the community that we want :)
MSmits: lol
AnalPainGiver: Lol, you better, or something else will come out hahahha
eulerscheZahl: i better go to bed before the urge to kick gets too strong. bye
MSmits: gn eulerscheZahl
AnalPainGiver: Ok,back to my ques, why my xp's are not increasing from solving clash problems
reCurse: Ah to be 12 again and find those things funny
Astrobytes: Oh, I go afk for a minute and get this to come back to
JBM: as eZ intended to say
JBM: there'x no xp for clash
AnalPainGiver: ok, then how to increase level?
JBM: with stuff that yields xp
AnalPainGiver: and increase ranking?
JBM: mostly puzzles
AnalPainGiver: ok
AnalPainGiver: And that I can find under which sectino?
AnalPainGiver: section*
JBM: but also arbitrary stuff (see xpachievement page in your profile)
MSmits: Astrobytes this is not a typical troll, he just doesnt know the boundaries of civil communication
Astrobytes: I know him already MSmits
MSmits: ah ok
AnalPainGiver: Astro, oh, so you wanted to kick me out because of my explicit explanation of my username, huh
Astrobytes: No?
JBM: ranking, on the other hand, is yet something completely6Wmostly orthogonal
Astrobytes: I'll kick you out if you want but I don't care about your username right now.
MSmits: no, it could be a hemorroid, your name was fine until it turned out it was that other thing
AnalPainGiver: Astro, if Codingame's policy allow me to make my favorite user_id, that I mostly use every I play game. Then, It should be under filteration while making civil communication.
Astrobytes: MSmits: haemorrhoid! Dammit
MSmits: this time i didnt paste it off google
MSmits: google now thinks i have issues btw
AnalPainGiver: oh, you were referring to MSmits comment. Yeah, it's gross.. lol
Astrobytes: Hema/Haema is US/British
MSmits: ah
MSmits: yeah Brits always want their words longer
JBM: i thought americans wanted their words more pedantic
AnalPainGiver: It makes them sound like a gentleman...lol
Astrobytes: Hematoma/Haematoma, hematemesis/haematemesis etc
AnalPainGiver: I was having conversation with my teammate, she calls "Addi" for "Address".
zhoubou: Astrobytes I want to thank you and everyone else for pushing me through. I scored more than 40k on Code vs Zombies :D
zhoubou: My solution is simple and linear. I rewrote my entire code and found a lot of bugs in my sim.
Astrobytes: Ahhh well done zhoubou!
zhoubou: Thank you :)
Astrobytes: What's next on your list?
zhoubou: Ughh. Haven't thought this through yet. Am thinking as we speak.
jacek: :soccer:
zhoubou: I thought CvZ will take me much more time
darkhorse64: reach 100<%
darkhorse64: 100%
zhoubou: Hmm. It might be too difficult for me
zhoubou: At this point at least
zhoubou: Unless someone has some easy hints :)
Astrobytes: Oh you didn't get 100%? Your work is not finished :D
zhoubou: I don't even know how to start attaining that
zhoubou: Unless I implement some rules, other than random movements. Or there might be an optimization I'm not seeing.
Astrobytes: darkhorse64: a bit off the current topic but do you know if leRenard wrote the Thor/GThor Othello program?
darkhorse64: You would not believe it. My search is random (OK it's a genetic algo). Astrobytes: I don't know. Do you know his name IRL ?
Astrobytes: Sylvain
jacek: Q.?
Astrobytes: Author of Thor is Sylvain Quin
Astrobytes: It was after old john mentioned it and that he worked with him back in the day, I came across the references to the old othello programs and
Astrobytes: *and wondered
jacek: and yet they are beaten by the booker
Astrobytes: leRenards first submit went to top and was almost a straight copy of his old othello program from 30 years ago so...
struct: Also "he" wrote the book format from what I understand
struct: and not a book
struct: http://cassio.free.fr/cassio/custom_install/database/FORMAT_WTHOR.TXT
Astrobytes: oui c'est correct
Astrobytes: I mean yes, correct
pb4: "leRenards first submit went to top and was almost a straight copy of his old othello program from 30 years ago so..."
pb4: How do you know that ?
struct: He said so pb4
Astrobytes: He told us at the time.
Astrobytes: As soon as it went live he submitted.
Astrobytes: And told us when we were amazed.
pb4: code 30 years old ?!
pb4: Damn
MSmits: it was cool
reCurse: Had to write things pretty well back then
reCurse: Shouldn't be a surprise
Astrobytes: old john and le renard both worked together, with Michael Buro and others no less
MSmits: It wasnt even that easy to beat renards bot. took a while
reCurse: Who is
Astrobytes: ?
pb4: Who is Michael Buro ?
jacek: logistello programmer
MSmits: are we doing jeopardy?
Astrobytes: https://www.chessprogramming.org/Michael_Buro
darkhorse64: and many seminal papers on Othello
Astrobytes: Yep.
MSmits: heh that pic looks exactly the same as our othello... of course it should, but it's still strange to see
jacek: btw. not long ago they discovered those 'patterns' in draught community
pb4: I see the "GLEM" paragraph on that wiki page
Astrobytes: tric trac also has worked on... is it checkers/draughts? for decades
reCurse: All these famous people in hiding
jacek: which is exactly what othello did then
Astrobytes: Hence his minimax mastery
MSmits: ahhh, well that explains why tric trac
MSmits: ... right
jacek: and yet no backgammon game
pb4: has anybody here applied something similar to that "GLEM" thing ?
Astrobytes: the tric trac game isn't straight-up backgammon afaik
pb4: More specifically this part : "but further provides a procedure for exploring the feature space able to discover new evaluation features in a computational feasible way."
reCurse: Ok Google, what is GLEM
jacek: pb4 thats n-tuple
jacek: more or less
pb4: I'll have to spend a bit more time reading about those n-tuples
Astrobytes: read up on MPC too
pb4: Ok Google, what is MPC :D
MSmits: to me it seems a lot like a NN, except you're giving it less freedom about the features it analyses
struct: multi probcut
Astrobytes: Sorry pb4, multi-probcut
MSmits: multiprob cut is like cutting of branches based on probability
MSmits: to search deeper in ab search
MSmits: if i understand it correctly
Astrobytes: You do the probablities offline
MSmits: yeah the constants
MSmits: and there are many
Astrobytes: *probabilities
Astrobytes: Yep
MSmits: old john pushing me in D&B
reCurse: multi-pb4cut
Astrobytes: :)
Astrobytes: he's online MSmits?
MSmits: not currently, but i see him a lot in my last battles, since 2 hrs ago
Astrobytes: ah ok
MSmits: after i submitted
MSmits: i think he won a lot vs me, because my bot was 100% deterministic before
MSmits: it's got time based random seed now
Astrobytes: Aha. Throwing him off the scent.
reCurse: Gross
MSmits: yeah i had a fixed seed before which won 80% vs remi, the time based seed wins only 65%
MSmits: remi is very deterministic also... so we were mostly playing the same games with both using fixed seeds
reCurse: This is why friends don't let friends make deterministic games
MSmits: right :)
Astrobytes: heh
MSmits: this is a very underestimated problem I think and one of the causes of RPS effects on leaderboards
reCurse: Underestimated? I keep whining about it for years
MSmits: yes, but you are an outlier
MSmits: :P
reCurse: Story of my life :'(
MSmits: i noticed it mostly because of my book experimentation
MSmits: you keep checking games to see what your bot does wrong and suddenly you recognize the same game over and over
MSmits: games having inherent randomness (in the referee) solves this too
reCurse: Random initial state is best
Astrobytes: I think reCurse is more getting at starting positions etc rather than straight up random-ass games
struct: I was gonna finish Amazons, but its also bookable to some extent
MSmits: i agree reCurse, it is better for bot games. But people like those fixed opening state boardgames, because they are familiar and well studied
Astrobytes: Not completely iirc. And it's worth it I think.
reCurse: That doesn't prevent it from being randomized with openings
Astrobytes: @Struct
reCurse: Look at TCEC, random forced openings
reCurse: Otherwise you'd get the same game every time
MSmits: TCEC?
reCurse: Computer chess championship
MSmits: ah ok
reCurse: And it's applicable to any game with a fixed initial state
MSmits: well players can do the randomness themselves
MSmits: if the openings have equal value
reCurse: Do pairings, force random moves, there you go
reCurse: Extra points if you force good random moves
reCurse: It's trivial
reCurse: Besides if you want to compare the human elemen
MSmits: yeah i understand the organization can force this, but the players can force it too
reCurse: If you lose doing an opening you won't do it over and over again
MSmits: by putting a random component into their bot
reCurse: Unless your name is Puppey or something
MSmits: dont understand the puppey reference :0
reCurse: Throwaway joke
reCurse: Don't bother explain it
MSmits: kk
Astrobytes: dota?
reCurse: Yes
MSmits: I played dota2
Astrobytes: Had to google but don't get the joke, haven't followed it much.
MSmits: I think 200 games total or so
reCurse: You know what they say about explaining jokes
Astrobytes: Yep.
MSmits: it's the best way to learn humor!
Astrobytes: Not always.
MSmits: that was a joke
MSmits: want me to explain?
Astrobytes: 'in-jokes' make much less sense.
MSmits: in jokes?
MSmits: inside joke?
Astrobytes: fml
Astrobytes: yes
MSmits: ah ok
MSmits: yeah those are not explainable
MSmits: until you explain the entire context
MSmits: by that point the person is inside
Astrobytes: getting back to not learning from losing opening moves
Astrobytes: ...
MSmits: what i've noticed about the deterministic games we have is not just that they are bookable
MSmits: even if the best bots use a random component
MSmits: the tree of "good moves' is just very small
Astrobytes: narrow lines of play, yeah you've mentioned that a lot
MSmits: yeah it's what i've noticed
MSmits: so random openings helps if a game has many options
MSmits: like chess
MSmits: othello doesnt have that many
MSmits: unless you force like.. the first 5 plies or so
MSmits: maybe more
Astrobytes: Not sure if anyone has experimented with that
MSmits: nah i think the game is just too simple to be interesting for those contests
MSmits: it used to be more interesting when the field was young i suppose
MSmits: and computers were slower
MSmits: about amazons, doesnt this game have huge branching?
struct: yes
MSmits: that makes it unbookable by definition, like dots and boxes
MSmits: it would only be bookable if someone had a perfectly deterministic bot and someone else bothered to make use of that
struct: "Also, the huge branching factor, over 1000 available legal moves during most of the opening, makes selecting good moves difficult. Still, one or two bad moves in the opening can doom a player to passive play and an almost certain loss"
MSmits: thats a waste of time i think
MSmits: might be worth making a multi of
Astrobytes: It was my suggestion a long time ago
MSmits: good one
MSmits: yeah i remember you saying it multiple times even
MSmits: thats when i learned about it
Astrobytes: You remembered something someone said? You been on the nootropics? :D
MSmits: hey i try :P
Astrobytes: hehehe
reCurse: Cute old couple aww
MSmits: hey, dont you need to go for a walk by now
Astrobytes: arrr shut it grandad
reCurse: Already did
Astrobytes: lol
MSmits: aw ok
ISAAColl: hay papi
MSmits: allright i need to go grade these christmas cards now. Want to get at least that much done before sleep
Astrobytes: Ah the python turtle ones. Enjoy :)
MSmits: yes, thanks and gn
Astrobytes: gn MSmits
zhoubou: I'm reading up on genetic algorithms. I'm wondering what the genes might hold. I'm just doing random movements, so I can only select between random movements. How can I represent that in a gene?
jacek: MSmits there are othello 'xots'
jacek: starting balanced openings with 8 plies
jacek: http://berg.earthlingz.de/xot/aboutxot.php?lang=en
k4ng0u: zhoubou you can take an angle and a distance for instance but it highly depends on what you try to solve
Astrobytes: CvZ k4ng0u
Astrobytes: So just x and y in this case
Astrobytes: @zhoubou
darkhorse64: or radius and angle
Astrobytes: jacek: interesting, I hadn't seen that
Astrobytes: darkhorse64: oh indeed
zhoubou: Hmm. So a gene would be (x, y)? Astrobytes
Astrobytes: zhoubou: see darkhorse's reply
zhoubou: Or (r, theta), yeah
zhoubou: But that gene is only good for a single turn.
Astrobytes: Well you evolve your population over several turns
Astrobytes: You keep good genes in a pool, mutate them, and cross over the best ones. Rinse and repeat.
zhoubou: I know the gist of it, yeah.
jacek: you do it online?
zhoubou: What I'm not understanding is the representation of the gene/chromosome for this specific problem
zhoubou: jacek How can I do it offline? Unless I cheat.
Astrobytes: Just because the validators are known doesn't technically make it cheating
Astrobytes: kovi may disagree
jacek: mhm
darkhorse64: It can be done online. My score is online
AI_says_hi: darkhorse64 wait how do you train online?
darkhorse64: ofc my code is C++ for maximum speed
darkhorse64: I search online
Astrobytes: Simulate multiple turns ahead and score your solutions, choose the best one
darkhorse64: I simulate till the end
AI_says_hi: Oh so you find suffiently good "genes" in that short time? impressive
Astrobytes: zhoubou: your chromosome/solution is just an n sized array of moves and a score
darkhorse64: Yes but the key is to keep on refining after the first turn, 1s is lot of time but 50 ms matter also
struct: My non hardcoded version was just MC
struct: managed around 650k
darkhorse64: The sim is not complicated so you can run a lot of them
zhoubou: Well then I need to optimize it o.o darkhorse64
zhoubou: I see Astrobytes
Astrobytes: So you can mutate solutions/chromosomes and crossover
struct: My performance was not great there
struct: I have like 50k games simulated on the worst test case
struct: Hoard
zhoubou: Online?
struct: yes
zhoubou: Wow..
struct: I mean
struct: 1 turn = 50k games
zhoubou: For me, it times out after 50 or so games on some test cases
zhoubou: Lol
struct: language?
zhoubou: Python
zhoubou: Around 500 for Cross (10th) test case
zhoubou: Actually around 100 for Hoard
struct: ok my number 40k was wrong
struct: is less
zhoubou: But it's much more than 100 I presume
struct: its around 25k on turn 2
zhoubou: Yeah..
zhoubou: So I have a big problem
struct: Well im using c+p+
zhoubou: Yeah, I don't want to switch. Not yet...
darkhorse64: Searching with Python is ... ambitious. I go 1.2 M simulations steps for test case 10 first turns
reCurse: "A game may last theoretically up to 209 moves but, in practice, a game is generally decided before 100 moves are played."
darkhorse64: turn
reCurse: What is a move, 1 or 2 players
darkhorse64: A ply, a move by one player
jacek: as it should be
jacek: why chess folks cant understand this
reCurse: My current average is 150, don't know if it's a good or a bad sign
darkhorse64: Slowly building up your advantage ?
reCurse: Or having no clue how to play
RoboStac: this for bt?
reCurse: Yes
jacek: msot games are less than 100 plies
reCurse: Yes that's what is written
jacek: from experience
RoboStac: when I start training the games get longer, then start to come down and are mostly below 100
darkhorse64: You should lose very fast then. This is the issue with bt. Search happens on all the board but only far advanced pawns really matters. The winning combination get lost in the fog
jacek: huh? when i start training my games are very short. one wrong moves and bah!
reCurse: Problem is I have zero clue what is good play
jacek: good play is when you winning :V
reCurse: I did get that longer then shorter part
reCurse: But 150 still seems like a lot
darkhorse64: Search for Dan Troyka papers on the strategic aspects of bt. There are good tips for humans
reCurse: Oh cool, thanks for the tip
reCurse: Then again it's early on the training
jacek: yeah, searching by 'breakthrough' sucks
reCurse: ETA is currently around 12 hours to get decent number of iterations
reCurse: sigh
struct: How many games do you play on that time?
reCurse: I'm at around 25k games for 30 minutes right now
darkhorse64: You guys are making me feel like a caveman with my mcts
jacek: mcts is so last decade
reCurse: Would be 5 times faster if hardware was accessible
darkhorse64: I am so last decade
struct: Only way to get parts is to buy prebuilts it seems
reCurse: I don't want to overspend over msrp either
reCurse: Against my religion
struct: Yeah, they ask for 50% more
struct: its way too much
reCurse: Anyone has experience setuping up cloud instances? :P
Illedan: What are you talking about?
Illedan: *game
reCurse: bt
Illedan: bt?
reCurse: bt
RoboStac: 25k per 30 mins should let you train something fairly quickly (it's quite a bit higher than I get and my current BT bot was less than a days training)
Astrobytes: Breakthrough
darkhorse64: Breakthrough
darkhorse64: damn ankles
Astrobytes: hahahaha
reCurse: Yeah that's hoping my hyperparameters are right
reCurse: If not then I have to start over
RoboStac: though obviously theres a lot more too it than just games
RoboStac: yeah
reCurse: I dunno maybe uttt spoiled me
jacek: mods everywhere, i feel so insignificant
Wontonimo: you can test or search hyperparameters
reCurse: Sure, still need the compute to go at it fast
Wontonimo: +1
reCurse: I'm quite impatient when it comes to training
reCurse: Need to write a dashboard with cute widgets to pace myself
reCurse: Live game would be nice
Astrobytes: You've not done that yet?
Astrobytes: I thought you would've baked that into your framework tbh
reCurse: No, framework requires gpu to run
reCurse: Wasting precious resources for training
reCurse: Even having CG open is a compromise
Astrobytes: gotcha
reCurse: Though I'm not sure how much is shared between cuda and 3d rendering
ChuFta: guys please join https://www.codingame.com/clashofcode/clash/1545531de3dbd549841a877ac29e9a4804c05a2
reCurse: But having a viewer open I notice an immediate dip in perf metrics
reCurse: Yay pixi
ChuFta: is there a way to start "shortest mode" clash of code which is publice??
jacek: no
Astrobytes: Sure. It's called Code Golf in the Compete section. No time limit but it is public.
Astrobytes: But don't compete, just try to have fun instead.
Astrobytes: :salt:
Astrobytes: reCurse: your normal 'battlestation' is built on openGL or dx or?
reCurse: opengl
reCurse: Problem is it's realtime instead of smart updates like an oldschool gui
reCurse: So it's running like a game, 120fps
Astrobytes: Yeah it's gonna hit the gpu
reCurse: Doesn't matter at all until I need 100% of my gpu
struct: Cant you limit fps? or it doesnt matter
Astrobytes: which is gonna be quite frequent now by the sounds of things
reCurse: Sure but it's still going to hit the gpu
Astrobytes: c# gui and text display for you then
reCurse: I think I'd rather do web at this point *shudder*
Astrobytes: :D
reCurse: I am so burned out with c# gui
reCurse: A real shit show
Astrobytes: Time for VB
Astrobytes: But seriously, it's not nice.
reCurse: Still not sure why UI is such a mess
reCurse: Ok just kidding, I know why. Just wondering with all the strides made in other areas.
Astrobytes: Guess they just gave in to web-based/pthon/java
reCurse: 3d rendering in 20 years is a herculean effort
Astrobytes: *python
reCurse: UI in 20 years?
Astrobytes: mmm
reCurse: Maybe effort is not the right word
reCurse: Progress
reCurse: There
reCurse: 20 years later and many times I'd rather have a html page
Astrobytes: Of course there's not just the whole experience of actually having to develop UI, but the fact the the UX/HCI field can't really keep up
Astrobytes: Some of the traditional 'rules' still apply but with ever changing mobile device incarnations it becomes a real mess.
Astrobytes: imo
Illedan: Agreed on C# UI mess.. MVVM -.-
Astrobytes: I'm not really sure a lot of people coding any kind of UI really understood anything from HCI
reCurse: The pre-web era windows folks were godlike tbh
reCurse: Underappreciated how much they understood HCI
Astrobytes: Yes, you're not wrong. There's a reason for the market dominance there.
Astrobytes: "I'm not really sure a lot of people coding any kind of UI" + nowadays btw
reCurse: I wouldn't mind MVVM so much if it wasn't for that goddamn abomination called WPF
reCurse: You can put a spinning cube in a dropdown but good luck if you want to have a tab that works as expected.
reCurse: Or something, forgot my examples, I exiled that land years ago
Illedan: Haha, I just left that land
Illedan: Now I'm in React
Illedan: seems far better
Astrobytes: But... but... WPF is so.. easy to use and quick to develop with :P
Illedan: Bloated
Astrobytes: No shit.
reCurse: Obscure, overengineered
reCurse: Unmaintained
Illedan: Might work in Xamarin if the Functional approach gets good
reCurse: Just stay clear from Microsoft UI frameworks
reCurse: inb4 just stay away from microsoft lul
Illedan: I started using Avalonia for my personal projects
Astrobytes: What! MFC + ATL FTW!!!!
reCurse: Avalonia looked like abandonware
Illedan: Indeed, but it works on my windows and my mac :shrug:
Astrobytes: Hm, hadn't seen avalonia
reCurse: I'd just like a minimalist ui toolkit for web
reCurse: But form trumps function these days
Illedan: C# webapp?
reCurse: I have my doubts this is going to go well
Astrobytes: the webasm thing?
reCurse: Different things
Illedan: Blazor
Astrobytes: yeah I know Illedan, was just wondering which part of it reCurse was doubtful about
reCurse: Very limited widgets it looks like
reCurse: Sorry didn't know blazor
Astrobytes: It's been a hot topic for some time, you really didn't hear of it?
reCurse: I don't follow the web scene that much
Astrobytes: Neither do I but some things you pick up through osmosis.
reCurse: shrug
samples32: What is the easiest practice game?
struct: Tron might be one of the easier ones
samples32: Thanks
samples32: How am i expos to tell if im about to run into my light beam. P.S. Sorry for the misspelling
Astrobytes: You have to keep a record of the cells you have visited
samples32: Thank you Astrobytes.
Jasperr: Is there a way to go from a clash results page to the clash contribution?
samples32: I think you have to go to the contribute page.
Jasperr: that shows all contributions
Jasperr: seems like what I need is not available
struct: Search it here
struct: http://eulerschezahl.herokuapp.com/codingame/puzzles/
Jasperr: if only I would remember the name of the clash
Astrobytes: if you remember keywords you can use them
Astrobytes: anything from the description
Jasperr: oooh thats very useful
Jasperr: thanks struct
samples32: in the top left corner their is a dropdown just for that.
struct: Today I got this clash
struct: https://www.codingame.com/contribute/view/491041a64dbe793c2fb39954589eb66138b3
struct: Dont know how it was approved, the statement makes no sense
Jasperr: I just had this one
Jasperr: https://www.codingame.com/contribute/view/5812f84ec1ca91bb0c6ed404a3e78d4734b4
Jasperr: so much broken english I had to find it to fix it
Astrobytes: wtf struct
struct: I understood that one
struct: astro yeah, either someone edited statement
struct: or I dont know
struct: "With only numbers 1, get the highest score using only the operations +, *"
struct: 1+1 = 2
struct: answer for 1 should be 1
Astrobytes: Yeah it's not clear at all
struct: (╯°□°）╯︵ ┻━┻
Jasperr: lol
Astrobytes: since you have only 1 1
reCurse: Quality
AI_says_hi: struct it's kinda wtf but I think what they mean is you have N ones and you can + and * until only one number is left
struct: ah I understand now
struct: Makes sense now
Astrobytes: Yes but how does "With only numbers 1, get the highest score using only the operations +, *"
Astrobytes: actually tell you that?
AI_says_hi: Thats what I havemy crystal abll for :p
AI_says_hi: ball*
Astrobytes: heh
struct: 90% of the clash is to understand the statement
struct: 95%
AI_says_hi: ^ TRUE
Astrobytes: Hey, no one cares about the 5%...
Chorizo: hi i wan't to what do i have to put in the box solution when i create a new contribution
Chorizo: the code ?
struct: yeah
Astrobytes: The code to solve it yes
Chorizo: ok thx
Chorizo: https://www.codingame.com/contribute/view/6101e1551ee561c3928320d8537d81b4a182
JudgeAL: https://www.codingame.com/contribute/view/61029c19e127354559b93908d1dcfedac3ff
Westicles: JudgeAL, that is just terrible
JudgeAL: whhaaa
Jasperr: calm down satan lol
struct: Impossible to solve
bobth: my coc room getting this
bobth: An error occurred (#73): "Only 1 executor running at the same time for a test session".
bobth: slow for other people too
struct: Servers get slow around this time
struct: to update db or something
bobth: oof
MartinEduardo: Hi
MartinEduardo: anyone has solved the ascii art puzzle? maybe that person could tell me if it is hard or not that hard at all
struct: Might be one of the hardest out of the easy ones
MR3M3AT: Hellooooooooooooooooooooooooooo
MR3M3AT: Whats the best coding language?
RyanGilbert: uh
RyanGilbert: bad question
badAtCoding: what r u trying to do?
NguyenDuyPhu: I have completed all test cases of puzzle "Find happy number" but when I complete it it is only 80%. Where should I find the remaining test cases? : ((
ZarthaxX: you can't see them, unless you are level 20+
NguyenDuyPhu: thank you
ZarthaxX: yw :)
Olusola: Clash Of Code again
IOI2020: how to get more xp
