Chat:World/2022-05-30
4nthracene: JOIN https://www.codingame.com/clashofcode/clash/2430567a75199b01a21ef992590d5a56d8b4c1b
4nthracene: https://www.codingame.com/clashofcode/clash/2430567a75199b01a21ef992590d5a56d8b4c1b
5DN1L: please use #clash channel for clash invites, thanks. 4nthracene
jacek: i thought ddpg was only for continous actions?
waterproofsodium: gm
waterproofsodium: I see you survived the taco :)
gokubill: halo
gokubill: hi
gokubill: hi hi hi hi hi
5DN1L: don't spam, gokubill
gokubill: i not spam =]
gokubill: just say hi
5DN1L: i'll kick you out if you spam hi or anything else again
gokubill: =]]
derjack: :upside_down:
derjack: darn, where is your brother, Automaton2000?
Automaton2000: i would be happy to help
TINOUAINANI: [[=
waterproofsodium: taco?
waterproofsodium: rip
RGNBLZD: Hello world
HerkusP: print("Hello World")
Uljahn: print("Hello Automaton2000")
Automaton2000: but you still need to get this to work
feuxclenoir: fuck nigars i love hitler, shut the fuck up bitches www.pornhub.com www.goku.com.
5DN1L: banned
Yuhiii: lmfao
Yuhiii: how he can tell that here
icyee: what no grass does to a man
Yuhiii: it's so stupid
shaau: lol
Yuhiii: i'm 100% sure he is bad at all x)
Ayza: love it when someone says every swear word in the dictionary randomly
derjack: bloody hell
Dickson: i love how they ban him but not remove the hateful message lol
5DN1L: there isn't such a functionality...
Dickson: they should add it then haha
5DN1L: this chat is expected to be removed soon, so nope
waterproofsodium: :O
shaau: :scream:
Ajaiy: that was unexpected
TechTheAwesome: println!("Mamamia");
5DN1L: That was expected as it was mentioned a few months ago, though at that time no specific timeframe was mentioned
elie.f.asmar: So he was banned only from this chat or will he undergo disciplinary measures?
5DN1L: only from this chat
Dickson: he will be punished by law
TechTheAwesome: xD
Dickson: footnote law
TechTheAwesome: well an indefinite mute is harsh enough i think?
Dickson: no he should be executed
Dickson: like a program
derjack: like Automaton2000
TechTheAwesome: SEGFAULT
Automaton2000: okay i think i was thinking of something else
Dickson: can you not see the creator's answer to a clash of code?
Ajaiy: jacek: The image is ready, will upload in 1 minute. Get ready to embrace it
Ajaiy: Maybe I'll just wait until he/she responds or sth
lulzsimp: maybe after you've attempted the clash of code
lulzsimp: i havent tried
TechTheAwesome: nah there's no creator's answer, but anyone who join the clash can decide to show their own Dickson
TechTheAwesome: i guess seeing the correct answer just in case everyone else fails is a good feature
TechTheAwesome: but for now im pretty sure we don't have it
Dickson: ok I see, thanks for the reply @techtheawesome
Ajaiy: jacek: Are you online?
DomiKo: he isn't
DomiKo: but maybe I can help you :D
Ajaiy: No, it's not for help
DomiKo: ohh ok
derjack: what for
derjack: is this rick roll avatar?
Sephirious: I have a question that I hope someone knows the answer."If two numbers are equally close to zero, positive integer has to be considered closest to zero "
Sephirious: How can I achieve this rule ?
5DN1L: Have you tried anything?
Sephirious: My code can find the answers and etc.
derjack: if you have -5 and 5, then print 5
Sephirious: But output says -5 is smallerthan 5
Sephirious: https://www.codingame.com/ide/puzzle/temperatures
Sephirious: This is the question
Sephirious: I need to get output with negatives
Sephirious: If I convert them into positive
5DN1L: you may use another variable to store it if that helps
5DN1L: or test it using another conditional statement
Sephirious: Let me try thank you
5DN1L: np
JystoDussPipouPrime: hey ! when a clash finish, you can't continue it ?
5DN1L: you can't
Wild_Python: (python question)
Wild_Python: what is the diffence of chr() and ord()
5DN1L: chr: input is ascii value, output is character
5DN1L: ord: input is character, output is ascii value
Wild_Python: oh
nicola: They are opposites.
nicola: Reciprocal.
Ose: Hi every body
Prince6502: Hello, I am new please help me by
Prince6502: https://www.codingame.com/contribute/view/19221b88ef09467510e0f3ad5750ac40f74ce
Ose: You are not new you are on level 7!
Prince6502: i mean i contributed first time!
Prince6502: is it good or bad!
AbrahamJLR: Hi, I have a doubt. Can I change my name on the generated certificate at the legend level? (I don't want to show or full name when sharing it).
COOLRAPTOR08: hi
Dickson: Hello world
TechTheAwesome: im not too sure but it seems to me certificate name are one done thing
TechTheAwesome: @AbrahamJLR
Prince6502: https://www.codingame.com/contribute/view/192544b886e8f39b9633b77d7f5a26a9fe8d6
Prince6502: Suggets if any changes
antiwonto: [auto] Hey Youssif_Ekadi, here is a :taco: for loggin in today while it is quiet. You now have 1 tacos
waterproofsodium: nom
zakacat: Give :taco: zakacat
antiwonto: [auto] You used your taco giving ability for the day to discover that you cannot give tacos to yourself
zakacat: :expressionless:
waterproofsodium: :taco: waterproofsodium
antiwonto: [auto] You used your taco giving ability for the day to discover that you cannot give tacos to yourself
waterproofsodium: hm
waterproofsodium: this is unexpetec!!
jacek: oh my
Uljahn: :cookie: Automaton2000
Automaton2000: i want to do that?
wetones: why dont i get a taco?
struct: :taco: wetones
antiwonto: [auto] struct has awarded wetones 1 tacos. wetones now has 1 taco. struct now has 59 taco
jacek: so generous
Ajaiy: jacek: The image is done
Ajaiy: do you want to see it?
jacek: eeyup
Ajaiy: ok
Ajaiy: done
Ajaiy: you can refresh your screen
jacek: oh my, its beautiful
Ajaiy: It took time in pixel art
Ajaiy: And apparently I also rickroll everyone when I type
jacek: :tada:
Ajaiy: I guess I will change it every month or sth
Brawler_zx: I am afraid to ask how for "And apparently I also rickroll everyone when I type"
jacek: he changed his avatar. you need to refresh page
struct: just dont refresh ever
Ajaiy: lol
Ajaiy: but restarting your computer also works
Brawler_zx: restarting works but deleting system32 legendary
jacek: :upside_down:
antiwonto: [auto] ':upside_down:' was defined as ' ɥǝllo '
Wontonimo: hello other real humans like me
Wontonimo: hey jacek, did you read the super long discord msg I sent you?
jacek: yes
Wontonimo: after I wrote it, I got the feeling I was way off and you already do that
jacek: im still digesting it
Wontonimo: ah
jacek: yes
Wontonimo: it was a little long
jacek: i do that in my code
jacek: but cant get it working using nn libs
Wontonimo: i wouldn't try doing that with nn libs
Wontonimo: during learning
Wontonimo: the backprop won't work particularly well
Wontonimo: oh, you could do it with an embedding layer and a fixed number of input positions
Wontonimo: you want me to code something up on tensorflow for you?
jacek: no need, im not that desperate yet ~
jacek: i want to try to learn it by making
Wontonimo: for the backprop, I'd use the whole matrix. for prediction the addition/subtraction of rows is for speed
Westicles: discord msg? ew yuck
Wontonimo: if you were using tensorflow, you can use a one-hot + embedding layer to dynamically choose which rows to inject into the network
Wontonimo: https://www.tensorflow.org/api_docs/python/tf/gather
Wontonimo: then https://www.tensorflow.org/api_docs/python/tf/math/reduce_sum
jacek: like the one in checkers code
struct: these seem like avx operations
jacek: in the morning there was guy planning on making NN for spider attack
jacek: did he succeed?
Wontonimo: j4at ... let's look at the leaderboard ... just a sec
Wontonimo: he is still in the same place as yesterday
struct: If I could do a NN that would probably be the last game I would try
jacek: oh my
j4at: hehe
jacek: yeah, onitama is more obvious choice for NN
Wontonimo: HEY j4at !! :wave:
Wontonimo: how's it going on the NN ?
j4at: Didn't work on it today. :(
Wontonimo: what does your local off-line learning setup look like? Is it brutaltester generating data + DB to store + tensorflow/pytorch to turn data into learning ?
Ajaiy: What does NN mean though?
Wontonimo: or is it all online learning ... and if so how do you store versions etc
Wontonimo: NN
Wontonimo: what is NN
Wontonimo: where is antiwonto when you need it
Wontonimo: NN = Neural NEt\
jacek: eat :taco:
antiwonto: [auto] That taco command has moved to the #taco channel
j4at: training script - > opens brutal testers -> opens client.py -> uses sockets to communicate with the training script
Wontonimo: and is it online learning only or is there also experience replay or something else?
jacek: i bet most rl libs have replay buffer and/or target network by default
j4at: yeah it has replay memory but its pretty simple it stores last n states
j4at: there is some tricks that can be added but i'm too lazy
Wontonimo: says the guy who is creating 2 actors, 2 critics, and using state-of-the-art transformer network
Wontonimo: i don't think you can even say you are lazy. prioritizing your time maybe, not lazy
Wontonimo: I hope you destroy everyone in spider attack then write up a nice playground at https://www.codingame.com/learn to teach us all how to do it
Ajaiy: Co-op 100%
jacek: will it learn triple wind
j4at: I hope so
jacek: or will it exploit bugs like those hiders and seekers game bots
jacek: https://youtu.be/Lu56xVlZ40M?t=129
Ajaiy: Yeah i've seen that video
Ajaiy: the AI just beats the system
Ajaiy: Maybe in Spring Challenge you can control the other heros to do a triple wind or sth?
struct: I dont think that would work at all
Wontonimo: building a buggy game is poor programming, not a cause for celebration that an AI found an exploit
Wontonimo: I call that "marketing spin" not breakthrough tech.
jacek: why making unit tests when you can make ai to find bugs
1C0D3R: the devs didnt even know that bug existed
1C0D3R: ye exactly
1C0D3R: later there will probaly be AI testing bugs, asuming AI's arent the only game makers then
Wontonimo: also, it's been well documented that machine learning will exploit bugs, so nothing new there either.
jacek: gib me more computation power
struct: just make your code faster
j4at: jacek you might want to use vast.io it's the cheapest website to rent gpus
j4at: gpus server*
jacek: you use it?
j4at: yes
j4at: only for production tho, i do testing on my own computer
dot.nick: someone join event clash :( its just bots here
antiwonto: [auto] Hey dot.nick, here is a :taco: for loggin in today while it is quiet. You now have 2 tacos
dot.nick: https://www.codingame.com/clashofcode/clash/241259682d2d8682a9e20e417004eeb03a25dce
antiwonto: [auto] hey dot.nick dont paste those links here. Use the channel #clash
zakacat: Give :taco: antiwonto
antiwonto: [auto] Sorry but you can only award tacos once per day
zakacat: ur loss
j4at: That's a honest working bot. It doesn't accept bribe.
**j4at slaps antiwonto around a bit with a large fishbot
zakacat: :grinning:
zakacat: :fearful:
j4at: (╯°□°)╯︵ ┻━┻
ElJafar: kek
Darleanow: hey every1
BrandoCalderon: question, im coming around by the game
BrandoCalderon: and I noticed when I played CoC that JavaScript it's the only programming language to clash
BrandoCalderon: Can someone tellme what's going on?
Wontonimo: oh my, there are other languages other than javascript?
Wontonimo: jk, yeah, you can pick any of languages you want
Wontonimo: even php
**PatrickMcGinnisII :money_mouth:
Westicles: he means in the event
Westicles: only one language is allowed
Wontonimo: oh
Wontonimo: <loop> not me </loop> Wontonimo
PatrickMcGinnisII: +":taco:"
Wontonimo: oh, interesting. A language specific clash of code schedule
Wontonimo: next one is in 18 min and is shortest C
Wontonimo: will that actually get more users?
j4at: Wontonimo how to remove the timeout from brutaltester ?
j4at: \spider attack
j4at: I found this "gameManager.setTurnMaxTime(50);"
j4at: but changing the value doesn't really change anything
Wontonimo: why do you want to do this?
j4at: because my AI is slow :3
Wontonimo: right
Wontonimo: honestly, I don't know
Wontonimo: I havent ever tried
Wontonimo: it shouldn't be too difficult. i bet it is in the spider attack code, nto the brutal tester code
Wontonimo: anyone else know? struct , PatrickMcGinnisII ?
j4at: "gameManager.setTurnMaxTime(50);" is in spider attack
j4at: changed it but still getting timeout error
Wontonimo: :(
bobth: where is the language specific CoC event
PatrickMcGinnisII: I didn't do it for spiders, I'm looking tho
j4at: My training script waits for the 2 players to get inputs. Maybe that's why ? Should it get like inputs from first player that players send an action and then do smae with player 2 ?
j4at: Should it get like inputs from first player than player 1 sends an action and then do the same with player 2 ?*
j4at: then*
j4at: I'm sleepy I will just finish it tomorrow
Wontonimo: it's one then the other. they are not synchronous
Wontonimo: they are serial
j4at: that's explains xhy
j4at: why*
PatrickMcGinnisII: I had issues with the pattern matcher barfing on my input
PatrickMcGinnisII: the log file should tell you what it barfed on, i think it just deactivates player, then the gamemanager shows a winner
PatrickMcGinnisII: hmm
PatrickMcGinnisII: referee.java line 120 gameManager.setTurnMaxTime(50);
PatrickMcGinnisII: 102
j4at: Yeah I already tried that. The problem is not from that timeout i found out. Its just from my training script. It also timeout due to the 2000ms timeout in the start that's why increacsing the 50ms didn't change anything.
j4at: Fixed now ;)
j4at: Thank you anyways
PatrickMcGinnisII: try training an NN for clash of bots, I'm trying...but I'm new at it
PatrickMcGinnisII: it should be perfect for it actually
PatrickMcGinnisII: I put the 5x5 input matrix into a 50bit int
PatrickMcGinnisII: and there's only 10 possible outputs for each bot
PatrickMcGinnisII: so decisions aren't really fuzzy, they are fairly finite
Wontonimo: good luck !
Wontonimo: how are you training it?
j4at: yeah that seems a lot easier than spider attack
PatrickMcGinnisII: game doesn't give you any immediate feedback like HP and stuff
PatrickMcGinnisII: I'm trying to figure out how many botstates I actually need to store
PatrickMcGinnisII: so not therre yet
j4at: Are using RL?
PatrickMcGinnisII: max is currently 70Billion
PatrickMcGinnisII: reinforced, yes
PatrickMcGinnisII: using my own conpression techniques I can get just about 1% of that, lol
PatrickMcGinnisII: hmm, maybe it's just over .1%
PatrickMcGinnisII: so I figure, i can write early game avoidance movements... and late game should use the library I end up with
PatrickMcGinnisII: library should only need to be about 40Mb (uncompressed) ... how to get it under 80k, ihnfi
j4at: damn that's a lot
j4at: My sipder attack NN has +1m float16 weights.
PatrickMcGinnisII: lol, can almost weight every pixel on the board
PatrickMcGinnisII: i did a heatmap with the random spider spawn + movement... didn't show me anything i already didn't know
PatrickMcGinnisII: so GL
j4at: thank you ;)
ShadoDrago: how do I use code I am a bot
j4at: My NN has 12.5 million float16 parameters actually that's more than I thought :( If i remove 1 trick It will become 2m instead without really lowering size of the NN. And If i lower the size a little it will become 900k. That should be enought after compression but it will lower the quality of my NN alot.
antiwonto: [auto] Hey j4at, here is a :taco: for loggin in today while it is quiet. You now have 6 tacos
antiwonto: [auto] ':taco:' was defined as ' A currency to some, sustenance for others '
PatrickMcGinnisII: j4at can u remove the high bit w/o performace loss?
j4at: Even if i can that's like 1/16 compression
j4at: 15/16*
j4at: I'm starting to give up
PatrickMcGinnisII: There's also line length limits, i think it's 1k chunks
PatrickMcGinnisII: the real experts on shoe horning data into CG is def. not me
jacek: i do 1 weight 1 char, with reduced resolution
jacek: so realistically you can get to around 80k weights
PatrickMcGinnisII: I can get base53 conversion and then LZ into 999 char chunks, and reverse at startup ... but i know there's better ways depending on lang.
j4at: what is a realistic compression ratio that i can achieve ?
PatrickMcGinnisII: I think it's 20MB to 90k...but I've only gotten like 5MB to 40k, whatever that is
jacek: oO
j4at: damn
PatrickMcGinnisII: jacek knows the ratios
PatrickMcGinnisII: My libs aren't all that when i use 'em
PatrickMcGinnisII: ratio really depends on how differentiated it is
PatrickMcGinnisII: I think NN data is fairly dense ... so I'd guess 30 to 1...what u think jacek?
jacek: you mean you can get 2m weights into 100k code?
PatrickMcGinnisII: weights...no... sparse sets...sure
jacek: ahh. yeah
j4at: That will be a lot of work and I doubt that sparse sets will be that helpful in my case.
j4at: I will give up. I will train my 12.5m model and then train a smaller model to simulate the big one
PatrickMcGinnisII: wow research on compression algorithms on dense sets...the ratios are horrible
PatrickMcGinnisII: like max 4 to1
PatrickMcGinnisII: my kung fu is no good, soft woods
PatrickMcGinnisII: opening books are so different than NN