Chat:World/2022-07-04

From CG community
Revision as of 05:29, 5 July 2022 by Chat Log (talk | contribs) (Created page with "<img src=/a/45839106739408> CharonXD: bruh <img src=/a/75805763427049> China-xyh: /me <img src=/a/75805763427049> '''**China-xyh slaps around a bit with a large fishbot'''...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

CharonXD: bruh

China-xyh: /me

**China-xyh slaps around a bit with a large fishbot

j4at: yeah i find it confusing when puzzles have ^ in them I always think they mean xor

Darleanow: hello,

Darleanow: can someone explain why the list b changes ?

Darleanow: http://chat.codingame.com/pastebin/6d37366d-c337-44cc-a1c3-2ec6836ca15d

Darleanow: https://pastebin.com/ZTHaKF5q

Darleanow: b is basically [0,0] and then [1,0] then [1,1]

j4at: cg's pastebin doesn't work

Darleanow: yeah, theres a link right under :)

j4at: okay i just wokeup. I'm blind rn :)

Darleanow: xD

Darleanow: np fam :)

Darleanow: ok it's fixed

Darleanow: if b[i]=="0":

       tempres[i]="0"

Darleanow: this was the probleù

Darleanow: and i can't know why

Darleanow: so i replaced this by a counter that i set back to 0 at the end of iter

j4at: so changing tempres is changing b

Darleanow: and if i cahnge a value i +1 the counter

Darleanow: tempres=b

j4at: because they are the same list

Darleanow: tempres is a pointer to b

Darleanow: right ?

j4at: well, there is no pointers is python. But = never creates a copy of an object it just reference it

Darleanow: ik that there are no pointers, but is it considered like so ?

Darleanow: wdym by referencing it ?

Uljahn: https://www.codesansar.com/python-programming/aliasing.htm

Darleanow: thanks uljahn

Darleanow: :)

Darleanow: f=g[:]

Darleanow: something like this would word ?

Darleanow: copy() function xD

Uljahn: guess it makes a shallow copy, and you might need a deep copy sometimes

j4at: yeah that would work

j4at: there is .copy()

j4at: in this case atleast

Darleanow: deepcopy is work complex data structures

Darleanow: for*

j4at: from copy import deepcopy

j4at: f = deepcopy(g)

Darleanow: yeah ik :), but an exmaple of a complex data structure

j4at: but you don't need it in this case

Darleanow: i know ! i solved it already :p

j4at: Well, you also need to know when it is creating a new object or not, everything is a nobject in python including int, float, strings ...

j4at: http://chat.codingame.com/pastebin/618219db-578d-4cf8-90a0-7f71b0eac81c

aCat: hello

Darleanow: hey

aCat: just checking

j4at: A = 1 B = A B += 1

aCat: HOLD THE EXECTUTION!

j4at: print(A,B) # output: "1 2"

j4at: because += creates a new object

Darleanow: ok so a basic b+=0 would create a new object

j4at: well you don't need to really do that with ints because everything creates a new object

j4at: same case with all immutable objects

j4at: b = b+0; would create a new object; b = b*1 too

j4at: you really don't need to worry about int, strings, tuples at all

Darleanow: kk ty :)

j4at: also float and other types.. :)

j4at: np

Westunts: o/

Default avatar.png ThomasGottschalk: test test

aCat: test failed

aCat: chat removed :(

aCat: HOLD THE EXECUTION!

Default avatar.png SalvoUndMartina: penis

5DN1L: thanks for giving me the chance to ban you yet once again

InDeSchrank: you remember him ?

5DN1L: yes

InDeSchrank: did you miss him

5DN1L: no

InDeSchrank: why not he seems like a nice person

jacek: :upside_down:

j4at: https://www.codingame.com/share-replay/649050677

j4at: :upside_down:

j4at: needs to atleast pass struct

5DN1L: by reference or by value

j4at: huh ? both

jacek: he made 101

jacek: sub11k, noice

j4at: yeah :3

j4at: I wonder what the minimum time possiblee

jacek: at least 10359.64

j4at: bruh

jacek: you must defeat that rustian

j4at: My NN best score is about 11000. I use the best game that my NN that's why I got 10880. I can use whatever language to prove my dominance.

j4at: that my NN plays*

j4at: what language do you recommend ?

jacek: vb.net ;d

j4at: I will do :) After my NN finish training

derjack: :upside_down:

BlaiseEbuth: Eeyup

Darleanow: Hey everyone :)

Darleanow: guys, i don't understand

Darleanow: how people can solve thor in 50 chars

Darleanow: even with the golf

Mortis_666: they hardcoded the solutions

Darleanow: :o

j4at: 72 chars in Python :sunglasses:

Darleanow: 82 :p

Darleanow: i'll just harcode

Darleanow: i'll try **

Mortis_666: 61 chars in python with hardcode solution

j4at: It's kinda waste of time tho. I used 1 trick you don't to keep track of the horizontal position.

j4at: don't need*

Darleanow: Mortis, for the hardcoded one, u did something to get the case test, the u print a loc *n times?

j4at: validators are the same as test cases

j4at: you can just print lightX lightY initialTX initialTY

Mortis_666: the validators isnt same as the test case

j4at: ow

Darleanow: bruh, u litteraly have to guess it xD

j4at: yeha they are different

j4at: but you can read their names

j4at: 2nd is north and 1st is straight line

j4at: and 4th is optimal angle

Darleanow: indee

Darleanow: got the second

Darleanow: but can't send anymore kekw

Darleanow: hardcoding the first two is easy

Darleanow: but the last two is hard as hell :o

j4at: optimal angle is easy

Darleanow: not hard, looooong

j4at: nvm I misunderstood what optimal angle means :upside_down:

derjack: they change it periodically but sometimes validators dont have SW or NE

Darleanow: guys, where ca i find the weekly puzzle ?

5DN1L: https://www.codingame.com/events

Darleanow: thanks 5DN1L :)

Darleanow: glad to u see tho

5DN1L: it's now called a puzzle event, not a weekly puzzle any more

5DN1L: glad to see you too

Default avatar.png KiloKD: :nose_tone5:

Default avatar.png KiloKD: :drooling_face:

5DN1L: no spam please

BugKiller_328: hi

BugKiller_328: to implement bitboard, is it ok to use bitset?

BugKiller_328: bitset is enough fast as int ?

BugKiller_328: to implement bitboard

struct: no

struct: its not as fast

struct: but its ok to use I gues

Default avatar.png thiloxx: :stuck_out_tongue_closed_eyes:

j4at: any vb.net experts here ?

j4at: why Dim V As New List(Of List(Of(Integer,Integer))) From {{(1,1),(2,2)},{(1,1),(2,2)}} doesn't work ?? :/

BlaiseEbuth: :scream:

jacek: what have i done

5DN1L: you have successfully led somebody to learn vb.net

j4at: lmao

jacek: i should have told html

Default avatar.png elie.f.asmar: @j4at

Default avatar.png elie.f.asmar: For .NET 4.6.2 or lower, .NET Core 1.x, and .NET Standard 1.x you need to install the NuGet package System.ValueTuple

Default avatar.png elie.f.asmar: Install-Package "System.ValueTuple"

j4at: hehe ? I'm using cg I can't install anything

Default avatar.png noooooooooooooooooooooooo: Looking for puzzles that focus on OOP

Default avatar.png Aldoggen: I hope you're not working in python. you'd be focusing on POOP then

Default avatar.png noooooooooooooooooooooooo: Nah C++

Default avatar.png Nikanameu: you can code everything with OOP

Default avatar.png Nikanameu: object a + object b instead of a+b

Default avatar.png Aldoggen: for death first search I implemented a graph class

Default avatar.png Aldoggen: you could do that one

Default avatar.png noooooooooooooooooooooooo: Oh cool thanks guys :)

Default avatar.png noooooooooooooooooooooooo: I am new here

Default avatar.png Aldoggen: np

Default avatar.png carruca: hi

Default avatar.png noooooooooooooooooooooooo: Hello

Default avatar.png noooooooooooooooooooooooo: Are we able to upload our own puzzles?

struct: yes

Default avatar.png noooooooooooooooooooooooo: Oh

Default avatar.png noooooooooooooooooooooooo: So we make a suggestion for that and the community will decide if that will be added or not?

Default avatar.png Aldoggen: yes

Default avatar.png Aldoggen: you need to be a higher level to suggest things though

5DN1L: it's not just to suggest, but to actually create the puzzle itself

Default avatar.png Aldoggen: right

Default avatar.png noooooooooooooooooooooooo: Ahh that is so cool

Default avatar.png Aldoggen: you have full creative control, except the community won't approve it if it isn't good enough of course

5DN1L: well... :thinking:

5DN1L: let's say there are unfortunate instances

Default avatar.png noooooooooooooooooooooooo: I actually really like that idea as I have seen other sites where they just let people post w/e they want and it just ends up with a lot of low quality stuff

Default avatar.png Aldoggen: here you still get the occasional low-quality approved puzzle

Default avatar.png Aldoggen: but at least they're solvable

Default avatar.png noooooooooooooooooooooooo: Even if there are some it is definitely better to have a filter than not to have a filter

Default avatar.png noooooooooooooooooooooooo: true

Default avatar.png noooooooooooooooooooooooo: Also, there is no way to search the puzzles by their tags?

5DN1L: https://chadok.info/codingame/tags_list.html

Default avatar.png noooooooooooooooooooooooo: Oh are you one of the developers?

5DN1L: i'm just a moderator

Default avatar.png noooooooooooooooooooooooo: ahhh I see

jacek: of the chat. forever

Default avatar.png Aldoggen: wow. I'm beating literally everyone in wood 2 in line racing

Default avatar.png noooooooooooooooooooooooo: Thanks for the link it is super useful

Default avatar.png Aldoggen: lol jacek

5DN1L: jacek, i don't know you've become the spokesperson of CG :scream:

Default avatar.png Aldoggen: it's true isn't it?

Default avatar.png Aldoggen: for as long as the chat will remain, you'll be a chat mod

5DN1L: maybe i'll get fired before the chat dies

Default avatar.png Aldoggen: lol

Default avatar.png noooooooooooooooooooooooo: Haha will it be my fault xD

Default avatar.png Aldoggen: I'm rank 1 in wood 2, and the battles are only half done lo

5DN1L: yeah, let's blame it on the new guy :smirk:

Default avatar.png noooooooooooooooooooooooo: oh yeah blame me

Default avatar.png noooooooooooooooooooooooo: XD

Default avatar.png noooooooooooooooooooooooo: good idea

Default avatar.png noooooooooooooooooooooooo: :stuck_out_tongue_winking_eye:

Default avatar.png Ollle: guys in mad pod racing the race is just a black screen

5DN1L: try refreshing or try another browser?

jacek: free trial expired

Default avatar.png Ollle: wait there is a trial xD

Default avatar.png Aldoggen: lol jacek

Default avatar.png Licantf: hi, anybody can help with logic gates by b0n5a1? I'm trying it in scala or Java :D

jacek: hm?

Default avatar.png Aldoggen: I probably can

jacek: :tada: https://www.codingame.com/share-replay/649099870

Default avatar.png Licantf: @Aldoggen, how is your solution?

Unionize: Hi, i just did a Clash and even if my formula is correct that was not correct for the last test even if others players, in different langage, used the same formula. It was JS for me, is there someone who can explain me what's the difference in JS who caused this problem ?

Default avatar.png Aldoggen: your bots have a tendency to want to go downwards lol

Default avatar.png Aldoggen: Licantf I don't have one right now. show me your code and I'll help you

Default avatar.png Aldoggen: maybe do it in private chat

Default avatar.png Aldoggen: jacek euler keeps blowing you up lol

Default avatar.png Razerk: Hello I'm stucked in the puzzle (mad-pod-racing) where i need to change the power of the pod. I tried to put " + thrust" in the same consolelog than the x + y positions but not working. I have no clue to find the solution ..

5DN1L: what's the line of code you're using?

Default avatar.png Razerk: console.log(nextCheckpointX + ' ' + nextCheckpointY + ' 50' ); This ?

5DN1L: yes

5DN1L: it looks all right to me

5DN1L: what do you mean by not working?

5DN1L: not beating the boss?

5DN1L: or code error?

Default avatar.png Razerk: But now i want to add the "thrust"

jacek: no space between X and Y

5DN1L: oh

5DN1L: the number is the thrust

5DN1L: so in your example, thrust = 50

BlaiseEbuth: Fix your reading jacek

jacek: oh

Astrobytes: ezpz

Default avatar.png Aldoggen: though there is no space between Y and thrust

Default avatar.png Aldoggen: wait

5DN1L: there is

jacek: huh

Astrobytes: hehehe

5DN1L: you guys

Default avatar.png Aldoggen: I need to fix my reading too

5DN1L: all need glasses lol

BlaiseEbuth: Use my starter, go legend

5DN1L: :nerd:

Default avatar.png Razerk: console.log(nextCheckpointX + ' ' + nextCheckpointY + ' 50' + ' 80');


Default avatar.png Razerk: like this?

5DN1L: no

Default avatar.png Aldoggen: wow that's worse

5DN1L: you put in only one thrust

5DN1L: not two

Default avatar.png Aldoggen: the previous one was good

Default avatar.png Razerk: haaaaaaaa

Astrobytes: in your first example, thrust == 50

Default avatar.png Razerk: console.log(nextCheckpointX + ' ' + nextCheckpointY + '80');

Default avatar.png Razerk: This you meant?

Default avatar.png Aldoggen: that works too, except you need a space before the 80

Default avatar.png Razerk: the space don't work when I paste, I feel

Default avatar.png Aldoggen: it should

Default avatar.png Aldoggen: it did until now

Default avatar.png Razerk: Report me then haha, thanks for helping btw

Default avatar.png Aldoggen: is it working now?

Default avatar.png Razerk: Yeeees !

Default avatar.png Aldoggen: nice!

Astrobytes: :shrug:

Astrobytes: glad you got it working tho :)

Default avatar.png Razerk: Huge thanks to you 5DN1L

5DN1L: oh wow, np

Default avatar.png Razerk: yep thanks too Astro !

Default avatar.png Aldoggen: enjoy your spot in bronze

Astrobytes: you're welcome, I think ?

Astrobytes: :D

Default avatar.png Razerk: Thanks for your help Aldo, yep let's enjoy haha

5DN1L: not yet Bronze

Default avatar.png Razerk: Wood division omw

Default avatar.png Aldoggen: lol

Default avatar.png Aldoggen: np

BlaiseEbuth: ngl

wediaklup: chat is still here?

BlaiseEbuth: nope

Default avatar.png Aldoggen: nope it's gone

wediaklup: ah ok

Astrobytes: we are but a figment of your imagination

Default avatar.png Aldoggen: I was gonna say we executed some XSS but that works too

Astrobytes: :D

Default avatar.png Aldoggen: my line racing bot keeps doing dumb stuff

Astrobytes: in what way?

Default avatar.png Aldoggen: I can't articulate it. not even sure if it's dumb

Default avatar.png Aldoggen: it's probably better than I am already

Astrobytes: searching or just the floodfilling still?

Default avatar.png Aldoggen: okay now it just absolutely crushed these two randos https://www.codingame.com/replay/649114832

Default avatar.png Aldoggen: I said I don't know, it's probably just a combination of my eval and an unlucky seed

Astrobytes: there are some starting positions that are unavoidable losses iirc unless your opponent(s) do something insanely stupid

Default avatar.png Aldoggen: I can imagine

Default avatar.png Aldoggen: okay I think I should change the eval https://www.codingame.com/share-replay/649115192

Default avatar.png Nikanameu: i am here for the chat funeral when is it takin place

Default avatar.png Aldoggen: it's trying to annoy the enemy as much as it is trying to win now

Default avatar.png Aldoggen: need to win a bit more and annoy a bit less

Astrobytes: gotta go do some stuff, afk a few hours, good luck with the eval tweaking!

jacek: can Automaton2000 be annoyed

Automaton2000: so i can see that it is a very good idea

Default avatar.png Nikanameu: there is no annoying only good and bad move

Default avatar.png Aldoggen: Automaton2000 wants to be annoyed apparently

Automaton2000: let me take a look at this

Default avatar.png Nikanameu: Automaton2000

Automaton2000: do you have a real bot

Default avatar.png Aldoggen: and I can tab complete his name again

Default avatar.png Aldoggen: ayy this eval works a bit better https://www.codingame.com/share-replay/649115500

jacek: do submits work?

Default avatar.png Aldoggen: submits?

jacek: leaderboards seemed stopped

Default avatar.png Aldoggen: I went from wood 2 to silver in line racing today

Default avatar.png Aldoggen: seems to work

jacek: what about now

jacek: my submit stopped at 58% and i see in mad pod racing they are not moved for others

Default avatar.png Aldoggen: I'm submitting

Default avatar.png Aldoggen: it's pretty slow, but my battle progress is going up

Default avatar.png Aldoggen: 9%

jacek: green circle has leagues and repushed AIs from the challenge :tada:

jacek: so maybe its just slow

struct: nice im silver...

struct: :p

struct: I guess something broke on reusing it

jacek: still at 58% ;d

struct: The repushed part didnt work did itjacek?

brend0g: anyone wanna spunk on me?

brend0g: :grinning:

jacek: struct servers went green mode

Default avatar.png Aldoggen: lol

Maximoo62: brend0g... what the actual frick

Maximoo62: not cool bro

Default avatar.png Aldoggen: wow now I'm stuck at 9%

wediaklup: how do i explicitly copy an object in python?

Default avatar.png Aldoggen: deep copy or shallow?

eulerscheZahl: json.loads(json.dumps(object))

Default avatar.png Aldoggen: also, what kind of object?

Default avatar.png Aldoggen: lol

BlaiseEbuth: ngl

Scarfield: lol the leaderboard for green circle looks broken

wediaklup: i have a class and i want to make a copy of my object

Default avatar.png Aldoggen: wediaklup you could make a constructor for your class that accepts an object of your class

Default avatar.png Aldoggen: I don't remember if that works

wediaklup: that should work i think

Default avatar.png Aldoggen: you can only have one constructor, so it involves type testing your arguments or something

Default avatar.png Aldoggen: I'd just make a class method or static method that does it

Default avatar.png Aldoggen: Scarfield how is it broken?

BlaiseEbuth: o/ Breakfield

Scarfield: players from legend during the contest are now in silver.

Scarfield: oi blaise

Default avatar.png Aldoggen: I've heard something about that being the highest rank currently

Default avatar.png Aldoggen: yeah, don't know what's up with that lol

BlaiseEbuth: Where do you see that Blindfield? I saw it all normal

BlaiseEbuth: *see

Scarfield: https://www.codingame.com/multiplayer/bot-programming/green-circle/leaderboard

BlaiseEbuth: Ah yes. The legend league were drastically reduced. My bad hadn't scrolled

BugKiller_328: hi,

wediaklup: hi

BugKiller_328: minimax can be used if game will be more than 2 players ?

jacek: then it is n-max

MSmits: contest winners of green circle are in silver rank 1 and 2 with 173 people above them

Default avatar.png Aldoggen: look up minimax for more than two players BugKiller_328

jacek: or max-n :thinking:

BlaiseEbuth: With some adaptation and concessions yes

BlaiseEbuth: A paranoid

Scarfield: i suggest to move my bot from silver to legend, to even out the mistake, should be fair

5DN1L: maybe because the bots were submitted before the import, and the original bots won't be imported in such cases?

struct: I suggest mine to be promoted to legend

struct: since its above legend bots

struct: .9

struct: :)

BugKiller_328: how are you picking the minimax VS MCTS based on the game?

struct: I always use mcts

MSmits: if you're talking about mcts with a random rollout, then it depends on the game. Really long games, or games that potentially dont end (except on turn limit) are not good with mcts

jacek: if you havent good eval use mcts

Scarfield: rule of thumb, if you can evaluate a game state reliably minimax, if you cant evaluate reliably mcts

MSmits: but there's mcts with early playout termination, where you just eval at some point, then minimax and mcts are both using eval,

jacek: evalfield

Scarfield: NNcek

eulerscheZahl: will jacek be able to beat his own boss?

struct: ah double exp

struct: nice

struct: boss is almost finishing the battles

struct: but mirrored battles are not up yet

wlesavo: oooh, nice, is this a bribe for us to forgot about chat delete? :smiley:

eulerscheZahl: boss finished, move me up

struct: oh I need to get a battle vs some1 that is on submit

struct: so I get promoted

struct: its how its working atm

struct: fight me jacek

eulerscheZahl: ah, interesting

eulerscheZahl: i'll sit it out

eulerscheZahl: everyone playing vs smits will be gone the next moment

eulerscheZahl: yay, gold

eulerscheZahl: python gold boss?

eulerscheZahl: doesn't look right

struct: I guess gold is old silver

struct: just my guess because we start at silver this time

struct: but we were supposed to be bronze

eulerscheZahl: i have the same impression

struct: yeap boss seems to weak

struct: I guess they will revert it

eulerscheZahl: jacek your boss sucks

Razovsky: man don't say that

eulerscheZahl: at least struct can make it to legend now

Razovsky: during the challenge it was horrible to beat jacek

eulerscheZahl: i'm trolling, it's not jacek's boss anymore

kovi: how some are already in legend?

BlaiseEbuth: They paid

struct: kovi people that didnt submit

struct: on this arena

struct: + broken gold boss

kovi: i c

eulerscheZahl: Q

kovi: err....what? from wood4->legend without matches?

BlaiseEbuth: smooth

Razovsky: yeah

Razovsky: easy

Razovsky: :P

eulerscheZahl: gold boss isn't broken. probably just the silver boss. which in return might make the silver boss the old bronze boss

eulerscheZahl: CG christmas

Default avatar.png DevMarcao: just curious, for the ones who are on high level on CodingGame, do you put your profile on linkedin or in your resume? Or it doesnt change anything for tech recruiters?

Default avatar.png Aldoggen: I've put it on all my resumes I've landed tech jobs with

Default avatar.png Nikanameu: westiclees try harass me private forum omg

Sheeesh---: wait

Sheeesh---: how can i get back the code i used in the last competition?

eulerscheZahl: last code: from contest reports

eulerscheZahl: previous ones: your local git repo

struct: boss is now js

struct: lol

darkhorse64: Will I get promoted to Legend, then put back in Gold or even Bronze and loot XP again :yum: ?

eulerscheZahl: thank got they let the old boss submit finish before replacing it

eulerscheZahl: large legend

struct: who is this boss o.o

darkhorse64: The new one is not jacek's. There is hope for an even larger league

[CG]Thibaud: so Julien made a mistake and took another Gold bot as the boss. We'll let it as it is now.

struct: legend will be larger than gold

darkhorse64: Hmm, The JS boss looks very strong

jacek: :no_mouth:

Default avatar.png Aldoggen: I guess they've put more work in this boss than in the game itself

Sheeesh---: the games in arena are only one side

Sheeesh---: like you fight a bot one time

struct: The game is fine

struct: and they do not need to put work on bosses

struct: Since they can choose from one of the players

Sheeesh---: not two times like in the competition

jacek: struct legend today or delete?

struct: I think it's safe

eulerscheZahl: and then CG realizes some other mistake and closes legend for the day to make a new one tomorrow

struct: lol

jacek: why they didnt do it at friday evening this time

Scarfield: why not just move the leaderboard as is, like they did the previous contests?

eulerscheZahl: because some players have already submitted again. too late for a 100% move

Viveli: hello there

struct: They just need to open a league above legend now

struct: and everything will be fixed

darkhorse64: The boss looks approximately like a 150th rank from the leaderboard

eulerscheZahl: the diamond league?

darkhorse64: maybe even less

jacek: jacek league

Scarfield: xD

darkhorse64: the imaginary league

Dapps: what a pleasant surprise. made it to legend without touching my code

eulerscheZahl: because they made a bug in boss selection

Dapps: oh.. it's not because i'm awesome

eulerscheZahl: sorry to crush your dreams

darkhorse64: Another blow in our self-esteem

5DN1L: On Discord: http://chat.codingame.com/pastebin/c7e3bc17-ebbf-4e48-888d-d0cd61fe7e91

5DN1L: oops

eulerscheZahl: and the new gold boss looks weaker than the old one

Scarfield: its almost like giving out gold medals to all participants in kids soccer

5DN1L: Illedan — @[CG] Thibaud that seems like the wrong gold boss? Too many will promote 😱

jacek: participation medals?

5DN1L: [CG] Thibaud — it's not the same but I trust Julien to have taken a proper one 🤔 it seems he did not

darkhorse64: Aren't we all kids ?

struct: its too late now

struct: legend already has 149 players

[CG]Thibaud: yep, too late. Will stay as it is

darkhorse64: :rolling_eyes:

Astrobytes: wtf

Scarfield: xD

BlaiseEbuth: :speak_no_evil:

darkhorse64: :money_mouth:

Astrobytes: :expressionless:

eulerscheZahl: are we golfing chat messages now?

jacek: oh my

Default avatar.png Aldoggen: w

Default avatar.png Aldoggen: o

Default avatar.png Aldoggen: w

Astrobytes: yes, express as much as possible with one emoji

darkhorse64: time is scarce, euler

Default avatar.png Aldoggen: :kissing_cat:

eulerscheZahl: :pizza:

BlaiseEbuth: Scarcefield

eulerscheZahl: (i'd like to have a pizza now)

Scarfield: keyboard buttons can only handle so many presses

Scarfield: HollandaisEbuth

BlaiseEbuth: :thinking:

darkhorse64: Boss is suddenly on fire, promotion is gone

Default avatar.png Aldoggen: we can perform puzzle of the week on the boss

Default avatar.png Aldoggen: or is it already over?

eulerscheZahl: still weaker than the contest boss

Astrobytes: there goes my motivation again

eulerscheZahl: the puzzle of the week is a different fire Aldoggen. without boss

BlaiseEbuth: Puzzle of the weak

Default avatar.png Aldoggen: well, I'll try to think of a better joke next time

Scarfield: its like a motivational poster; we are all legends

Astrobytes: Shame on you Aldoggen :P

Default avatar.png Aldoggen: if everyone is a legend, no one is

Default avatar.png Aldoggen: lol

BlaiseEbuth: ngl

struct: Ok now I can see that my bot was not that weak

Scarfield: why not just make it a shared 1st for everyone

struct: I must have gotten unlucky

Default avatar.png Aldoggen: I don't understand why promotion to gold has to take so long

struct: Because submits need to finish Aldoggen

darkhorse64: 230 on going

Default avatar.png Aldoggen: I... what?

Astrobytes: you made gold in tron Aldoggen?

Default avatar.png Aldoggen: yup

Astrobytes: nice one :) Promotions to gold and legend on big multis always take ages

Astrobytes: to er, "force you to take a break" :rofl:

Default avatar.png Aldoggen: yeah, really don't understand why lol

Default avatar.png Aldoggen: kills my motivation

JohnyDaison: How did I get to Legend though? It's just a simple heuristic bot...

Default avatar.png Aldoggen: I had ideas to improve it

Default avatar.png Aldoggen: JohnyDaison green circle is bugged

JohnyDaison: Not enough CODE_REVIEW, eh?

Default avatar.png Aldoggen: I guess

eulerscheZahl: wait until you reach CSB gold and have to rewrite your whole code

Astrobytes: ah yeah, that

eulerscheZahl: CBS = mad pod racing

Default avatar.png Aldoggen: I already did

eulerscheZahl: did it kill your motivation?

Default avatar.png Aldoggen: wasn't fun

Default avatar.png Aldoggen: a bit yeah

Default avatar.png Aldoggen: now I'm stuck at around 700 or 400 or so

Default avatar.png Aldoggen: somehow made it there with only heuristics

Astrobytes: you can get legend with heuristics too

Default avatar.png Aldoggen: I'll eventually try to :)

Default avatar.png Aldoggen: but legend games are different too right?

Astrobytes: only heuristics also works for fantastic bits legend too

Astrobytes: sorry, "broomstick flyers"

Westicles: Nikanameu, what's this now?

Scarfield: for CSB i think the max thrust is changed from gold to legend, but that is the only change iirc

eulerscheZahl: 100 -> 200

eulerscheZahl: not sure if the 75ms/turn is a legend thing or everywhere

Default avatar.png Aldoggen: pretty sure it's everywhere

Default avatar.png Aldoggen: at least it's not instant

Default avatar.png Aldoggen: that's what got me through the lower leagues, realizing that I should hold my thrust while my bot was turning

Scarfield: -3vel ?

Default avatar.png Aldoggen: somehow CG has learned that I do my bot programming in java and my puzzles in python

Default avatar.png Aldoggen: me gusta

Default avatar.png BoBot: Hello BoBot, http://chat.codingame.com/pastebin/a0d7cd93-eebf-488c-a6b4-a98d93d4e697

Default avatar.png Aldoggen: nice

Default avatar.png BoBot: oops

Default avatar.png Aldoggen: on top of that, those pastebin links are broken

jacek: oh my

5DN1L: Aldoggen you're speed-running on CG?

Default avatar.png Aldoggen: am I speedrunning?

Default avatar.png Aldoggen: I'm having fun at least

5DN1L: did you get to Level 17 in 2 weeks?

Default avatar.png Aldoggen: uh yep

Default avatar.png BoBot: was going to write that I got Legend in Green Circle somehow, though I was not close to it during the contest... so either they made the boss weaker or I got reaaally lucky lol

Default avatar.png Aldoggen: BoBot it's bugged

Default avatar.png BoBot: the game?

struct: but you will stay in legend

Default avatar.png Aldoggen: 5DN1L I spent a lot of time on this site. maybe a bit too much

Scarfield: they made a mistake on picked a weaker boss bot

darkhorse64: no bugged but not the same bot

5DN1L: Aldoggen your speed is amazing

Default avatar.png Aldoggen: <3

Default avatar.png BoBot: well lucky me :D

Default avatar.png BoBot: yea, Gold is smaller than Legend now

Default avatar.png BoBot: haha

Sheeesh---: and here i'am in wood 2

Sheeesh---: because i was late to the party

Sheeesh---: gotta go fast

mannouao: :joy:

AntoninLdn: bonjour je cherche un stage

Rikus: qquestion, are there any bots that use NN (DQN, AZ, or similar) on games that have 50ms turn limit?

Default avatar.png Aldoggen: moi aussi I guess

0xhotdawg: mais pour what?

Default avatar.png Aldoggen: pour travailler no?

struct: Rikus yes

Sheeesh---: RIkus : https://forum.codingame.com/t/green-circle-feedback-strategies/196574/5

struct: Check jacek profile, most of his bots are NNs

Rikus: awesome! Ty for the info to both!

struct: and there are a few more games that have NNs at the top

jacek: yes there are

BlaiseEbuth: oO

Sheeesh---: wood 1 boss is in rank 127

Sheeesh---: lol

Wontonimo: congrats Aldoggen on your speedrun up the ranks !

jacek: :taco: struct

Wontonimo: sorry, tacobot is still unplugged

Default avatar.png Aldoggen: Wontonimo lol thanks, I'm just having fun

jacek: fun? here? :unamused:

Wontonimo: i can't think of a better reason

Default avatar.png Aldoggen: lol jacek

Default avatar.png Aldoggen: gold league bots everyone https://www.codingame.com/replay/649173690

eulerscheZahl: maybe some language update broke them

Default avatar.png Aldoggen: 3 -> possibly 4 bots maybe broke them

jacek: more exp for league advancing :Tada:

Wontonimo: oh, i see you are also doing bot battles Aldoggen. very nice. I liked tron. Are you working on it now (it's called line racing)

Default avatar.png Aldoggen: I am not working on it

Default avatar.png Aldoggen: currently*

Default avatar.png Aldoggen: I just got promoted to gold and I'm looking at the battles it's doing

Wontonimo: what is your background? are you a CS student?

Default avatar.png Aldoggen: math student

Default avatar.png Aldoggen: I took some CS courses as elective courses though

Wontonimo: is programming part of the course work for math these days? Computations analysis or something?

Wontonimo: ah

Default avatar.png Aldoggen: and there were some basic cs courses already included in the math program

Default avatar.png Aldoggen: like a basic introduction to programming in python

Default avatar.png Aldoggen: I did plenty of programming in a variety of languages, even outside the CS classes

Default avatar.png Aldoggen: languages like R, matlab, python of course. I've dabbled a bit with julia.

Wontonimo: i did some matlab also. don't remember any of it

Wontonimo: gotta go. Nice meeting you Aldoggen :wave:

Wontonimo: later struct jacek

Default avatar.png Aldoggen: cya!

struct: cya Wontonimo

Shivzee: Neffex OP

Default avatar.png Aldoggen: I'm currently implementing flood fill in a rectangular grid by setting elements of a twodimensional arraylist (yes, this is java). any ideas for a faster implementation?

jacek: arraylist of objects?

jacek: or primitives

Default avatar.png Aldoggen: String, so objects

Default avatar.png Aldoggen: I guess I could make them char if that improves performance

Default avatar.png Aldoggen: they only have length 1

jacek: so 2d array of chars

jacek: so no unneeded boxing/unboxing

Default avatar.png Aldoggen: wow I guess that can help

Default avatar.png Aldoggen: ideally I would build a sim (or whatever you call it) and test it

Default avatar.png Aldoggen: thanks!

Default avatar.png boumagik: the pushes are slow as hell

Default avatar.png boumagik: @codingame

Default avatar.png Aldoggen: it's been like this for a few hours

BlaiseEbuth: Hell is not slow. Don't tell such lies about my business or I sew you for defamation. :angry:

Default avatar.png boumagik: Slow as a turtle *

Default avatar.png boumagik: :)

Sheeesh---: i mean

Sheeesh---: everyone is pushing

BlaiseEbuth: :)

Sheeesh---: soo it's normal to be slow

Default avatar.png boumagik: Is there a contest going on ?

Default avatar.png boumagik: I'm just trying to push my bot to legend on springchallenge 2021

5DN1L: 534 bots being computed in Green Circle

Default avatar.png boumagik: arf

Uljahn: yep, past contest being introduced as multiplayer

Default avatar.png boumagik: Also, can anyone confirm : in challenges introduced as multiplayer, bots you win or loose against could go up and now in ranking, and you could see an arrow up green, or down red indicating the recent tendency

Default avatar.png boumagik: now it looks like it is fixed and never moving (for the othe bots)

Default avatar.png boumagik: they seem to stay at their rank

Uljahn: maybe the rank change is lower than some tolerance

Uljahn: that's how trueskill works

Default avatar.png boumagik: yes, in contest I suppose

Default avatar.png boumagik: but in multiplayer outside of it, is it enabled ?

Uljahn: yes, but the more games you have, the less the rank change

Default avatar.png boumagik: ok, might as well be

Uljahn: http://cgstats.magusgeek.com/app/multi-spring-challenge-2021/boumagik bots around you are kinda old, they had a lot of matches so their rank is very stable, you need to get a bunch of matches with one of them to change its rank noticably i guess

Westicles: Someone needs to approve Takuzu Solver

Westicles: https://www.codingame.com/contribute/view/18957f95843f0fd886c743c6c40f4e14aee89

Default avatar.png boumagik: Uljahn ok thanks for the input

Westicles: We already have Sudoku, Kakuro, Shikaku, Futoshiki, Hitori, and Nurikabe so this is a big omission

jacek: :upside_down:

BlaiseEbuth: Approve it yourself

Westicles: already did

jacek: approve from your alts

BlaiseEbuth: I'm too slow

Westicles: never done that before, not going to start now

BlaiseEbuth: There's a first time for everything

Westicles: If I didn't do it when Abusing the System had 2 approvals I never will

Westicles: Hmmm, actually it is still sitting at 2-2 :thinking:

Default avatar.png JustnTime: Damn the arena goes so slow

Default avatar.png SatelliteMaster1337: hello

Default avatar.png SatelliteMaster1337: does anyone use ruby

struct: only for golf

Default avatar.png SatelliteMaster1337: yeah lol is it useful anymore fo ranything else?

jacek: maybe snooker

Default avatar.png SatelliteMaster1337: thx ;)

Default avatar.png HenroLST: can someone explain to me how to reconsturct a path from a BFS with coords? i tried using a map where each coordinate was the key with its parent coord as the element but it ran too slow

5DN1L: That sounds like the right way to do it, maybe something else is the issue?

Default avatar.png HenroLST: it worked on the simpler test case but on a test case with a longer puzzle it just didnt work

Default avatar.png HenroLST: Lemme recode it and try again

Default avatar.png Aldoggen: I can take a look at it in 10 mins HenroLST

Default avatar.png HenroLST: yay, thanks!

jacek: woah struct in legend

struct: omg I dont have to delete now

Default avatar.png HenroLST: lmao

struct: how can your submit and msmits be finished

struct: when mine was in gold before you

MSmits: I use 1 ms most turns...

struct: so do I

struct: maybe the templates are slow

MSmits: templates are never slow

struct: and take a while to compile

struct: I mean for compilation time

MSmits: sure, they dont recompile the bot for every fight do they?

MSmits: also, i too use templates

struct: Some say no some say yes

struct: an answer we will never know

struct: I guess its easy enough to test

struct: ill use C4 for it

jacek: i use 1ms at most too

jacek: surely a green contest

MSmits: you are going to break into their office with C$?

MSmits: C4?

MSmits: sounds a bit excessive

Astrobytes: :D

struct: I need to recover chat

Default avatar.png Aldoggen: I just did the network cabling puzzle

MSmits: ahh that one

MSmits: it's a lot easier than it seems

Default avatar.png Aldoggen: I hate how they tell you to search a minimum, but the minimum is not a minimum

Default avatar.png Aldoggen: it's not optimal

Default avatar.png Aldoggen: arghhh

Default avatar.png Aldoggen: sorry

jacek: hm?

BlaiseEbuth: oO

Astrobytes: Oo

Default avatar.png Aldoggen: so my first approach was to use least squares to find the optimal height for that horizontal power line

Default avatar.png Aldoggen: nope, doesn't pass some tests

BlaiseEbuth: overcomplicated

MSmits: isnt this just manhattan distance?

MSmits: I think you may just have misunderstood the problem

Default avatar.png Aldoggen: maybe

Default avatar.png Aldoggen: I think it should at least be better defined

jacek: you dont need to understand a problem to solve it

Default avatar.png Aldoggen: yeha apparently

jacek: :smirk:

Default avatar.png HenroLST: sheesh just make a AI to solve it

Default avatar.png Aldoggen: you would know everything about that huh jacek? ;)

jacek: :no_mouth:

Default avatar.png Aldoggen: I'm just gonna plop an NN on it next time :)

Default avatar.png HenroLST: easy

Default avatar.png Aldoggen: though I still don't understand how there can be manhattan distance involved

5DN1L: Aldoggen use NN to solve Onboarding first: https://www.codingame.com/training/easy/onboarding

MSmits: if they are just straight lines then manhattan and euclidean distance are the same

Default avatar.png Aldoggen: ah, that's what you mean. thought you meant proper 2D manhattan distance

Default avatar.png Aldoggen: okay you know what? I'm gonna learn about NN and implement a simple one to do the onboarding puzzle

5DN1L: nice

jacek: there is one in the solutions

jacek: that is if you done onboarding

5DN1L: yeah Aldoggen won't be the first person to do so

Default avatar.png Aldoggen: lol

Default avatar.png Aldoggen: I'm not trying to be the first

Default avatar.png elie.f.asmar: hello

Default avatar.png elie.f.asmar: is it possible to change contribution type from Classic Puzzle to clash? (Status = Pending)

struct: Im not sure actually

5DN1L: it looks like you have to make the choice initially and can't change afterwards? i'm not sure either

jacek: probably not

struct: yeah you cant

struct: ok its only compiled once

PatrickMcGinnisII: jello , it's the 4th of July... the day all us Yanks blow stuff up

Astrobytes: movie celebration day :tada:

Astrobytes: :P

PatrickMcGinnisII: I bought $400 worth of mortars

pluieciel: “Someone needs to approve Takuzu Solver”;is there something like Prolog here? I'd like to learn some logic programming if possible

PatrickMcGinnisII: going to party, total fireworks by all atendees will total over $4k

jacek: https://img-9gag-fun.9cache.com/photo/arRK20y_460s.jpg

jacek: no prolog ~

PatrickMcGinnisII: happy no taxes day, oh wait . . . c r a p

pluieciel: no clojure.core.logic either

PatrickMcGinnisII: mass shooting already today, mf

PatrickMcGinnisII: timeout test

pluieciel: 408 Request Timeout

Default avatar.png Aldoggen: my favourite response code is still 418

BlaiseEbuth: The teapot one?

Default avatar.png Aldoggen: yup

BlaiseEbuth: ^^

Default avatar.png Aldoggen: someone probably has made a smart teapot that always returns this

Default avatar.png Aldoggen: someone somewhere

darkhorse64: There is only one teapot

BlaiseEbuth: :scream:

Default avatar.png Aldoggen: can you prove that?

darkhorse64: I am the teapot. This is not debatable

jacek: cna you disprove that?

Default avatar.png Aldoggen: I'd say we have two teapots at home, but I'm afraid they're coffee thermoses

Default avatar.png Aldoggen: or whatever those things are called

Default avatar.png Wishbone: I'd say it's pretty easy to tell a teapot and a thermos apart.

Default avatar.png Aldoggen: can you prove that?

TimeEngineer: Jacek can I know which activation function you use for your NN ?

Default avatar.png Aldoggen: linear

Default avatar.png Aldoggen: (jk. that probably sucks)

Default avatar.png Wishbone: Teapot: http://cdn.shopify.com/s/files/1/0527/2213/9300/products/johan-och-nystrom-viva-scandinavia-classic-teapot.jpg?v=1624584843

Default avatar.png Wishbone: Thermos: https://cdn.shopify.com/s/files/1/0552/9731/0876/products/tms-o-sk2010mdb4_front_v2_1800x1800.jpg?v=1624516304

Default avatar.png Wishbone: Quite easy to tell apart, actually ;-)

struct: He probably uses ReLu

Default avatar.png Aldoggen: I was gonna make a joke about being a topologist and not seeing the difference, but they're topologically different

Default avatar.png Aldoggen: you beat me wishbone lol

Default avatar.png Wishbone: Hehe, nitpicking is one of my primary skills :-D

Default avatar.png Aldoggen: lol

jacek: TimeEngineer for hidden layers its relu

jacek: and linear for output in dqn

TimeEngineer: I tried to implement from scratch a NN to play tictactoe, but I don't know why it's so difficult

jacek: or tanh in mcts stuff

TimeEngineer: Maybe it's because Softmax + Q-learning

jacek: suprisingly tic tac toe requires plenty of layers and/or nodes to play perfectly

TimeEngineer: And GreenCircle not ?

TimeEngineer: Do you think that Softmax and Qlearning can match together ?

jacek: it doesnt play perfectly, does it?

TimeEngineer: Sorry I'm newbie in AI :sweat_smile:

jacek: but yeah, green circle is a little more complicated than tic tac toe ;v

Default avatar.png Aldoggen: just a little :)

MSmits: TimeEngineer i got TTT to 97% perfect with 1 hidden layer and something like 30 hidden nodes

jacek: hmm normally youd use linear output layer for Q-learning, as it mimics the Q-table

jacek: softmax thing is more for actor-critic and others

TimeEngineer: :thinking:

TimeEngineer: That makes sens

TimeEngineer: sense*

TimeEngineer: MSmits with which activation functions ?

MSmits: the thing is, NN's don't have to be perfect when you're playing boardgames. Generally you're using them as eval at the end of a selection phase and backpropagate the result, averaging it with many other results. Because it is averaged, it doesnt have to be perfect

MSmits: I think i used relu

MSmits: softmax for output i guess

TimeEngineer: Supervised training ?

MSmits: this is why ttt is not a great testbed. If you search deeper, you soon hit the end of the game and require no eval at all. You'll just be mcts-ing it

Westicles: PatrickMcGinnisII, I'm jealous mortars are awesome

MSmits: I did supervised yes. I labelled all reachable states with labels from minimax

jacek: value n-tuples network for tic-tac-toe, supervised from all positions, https://github.com/jdermont/tictactoe-ntuple

MSmits: yeah thats the same thing except training ntuples weights instead of nn weights

TimeEngineer: Also I noticed something, When I train a lot, my weights can overflow, is it normal ?

MSmits: what do you mean overflow?

PatrickMcGinnisII: :boom:

TimeEngineer: They become big

MSmits: mmh i did not have that issue

jacek: and you get nan?

PatrickMcGinnisII: :flip: :boom:

TimeEngineer: I don't use framework, that's why

MSmits: I didn't either

Westicles: I'm stuck in this stupid blue state where you can't even have sparklers

MSmits: well not at first

PatrickMcGinnisII: (╯°□°)╯︵ ┻━┻

Default avatar.png [AaZZz09_.-]: IAM A NEW

struct: why 432 (8*54) weights jacek?

Default avatar.png [AaZZz09_.-]: NEED SOME SHELP

MSmits: i dont mind sharing my code if i can find it TimeEngineer

jacek: neither do i, i have small learning rate and i clip errors to be abs max 1

struct: [AaZZz09_.-] turn caps lock off

jacek: struct row-of 3 has 27 states. x2 for player to move. there are 8 such rows in TTT

Default avatar.png [AaZZz09_.-]: L

TimeEngineer: But it seems to be possible when I see a dense layer, there are a lot of addition :thinking:

Default avatar.png Vicvicvic: hey guys new here, i am having trouble on wood2 for next_checkpoint_angle > 90 or next_checkpoint_angle < -90: , my pod just shaking right and left and runs slow

j4at: this is my search race actor's activation function

j4at: https://i.ibb.co/TKmfqDY/Screenshot-from-2022-07-04-22-02-52.png

struct: ah I think I understand jacek

jacek: linear sin?

j4at: yeah

jacek: weird flex but ok

j4at: Normally people use tanh for ddpg. But it has bias because not the same gradient in all points

jacek: Vicvicvic do you do as said in the comments?

j4at: so you need to use custom gradient

j4at: or just use linear sin

jacek: the NN is quite sinner

TimeEngineer: Ok, thanks all I will try to implement with your recommandation jacek, I really want a dqn :smile:

jacek: at least its something you know it works

jacek: depending on the implementation and hyperparameters, it always took around 10k games to start winning over 90% against random opponent

jacek: it was like 0 0 0 0 0... and sudden spike to 99

struct: http://chat.codingame.com/pastebin/292831d4-4660-453c-bfab-8483d08dd31b

struct: this is waht you need

struct: https://github.com/asdfjkl/neural_network_chess/blob/main/chapter_05/sup_network.py#L8

struct: :D

struct: more layers

jacek: for hexapawn might as well do hashtable

MSmits: I did this

struct: might as well hard code it by hand

MSmits: https://trinket.io/python/6cd0e1872a?outputOnly=true

jacek: TimeEngineer also (maybe its obvious), if next state is other player, I use -Qmax(next_state) as target

MSmits: hexapawn with simple q-learning (hash table)

jacek: (notice the minus)

TimeEngineer: Really ?

jacek: its the move of the opponent and you want to minimize it

TimeEngineer: I used the same NN for the opponent and my nextQmax is 2 step after

jacek: oh. it could be just implementation detail

TimeEngineer: Is it a mistake, if so, why ? :thinking

j4at: the jacek way is better

Default avatar.png Aldoggen: that's why he gets a league all for his own for green circle

Default avatar.png Aldoggen: just below the last wood tier

j4at: because its guarantees that the opponent plays the "best" move

j4at: also it normalizes the reward

j4at: (makes it have a mean of 0)

TimeEngineer: Ah ! i made a mistake in my implementation, I give a reward of -1 in my softmax nodes xD

TimeEngineer: Thanks for the tip !

j4at: Using nextQmax 2 steps after makes it kinda on-policy. But dqn is off-policy SARSA is on-policy

jacek: oh my, i had small bug in my clash of bots sim implementation

jacek: need to train again -.-

j4at: first time ? I always make bugs in my simulations that I don't complain anymore

jacek: thats why i copied green circle referee. any bugs there are also in arena

Default avatar.png Aldoggen: XD

Default avatar.png Aldoggen: which means they are features

jacek: probably it would exploit that infinite draw glitch

j4at: My search race training script spends 50% of the time on the simulation. It's written in python and very very not optimized at all.

Default avatar.png Aldoggen: oof

j4at: Maybe I should rewrite it in Cython

DomiKo: just use c++

DomiKo: Hello :D

struct: hi

jacek: then its something wrong. my NN stuff take over 90% time

DomiKo: I agree with jacek

struct: my NN takes 0% time

DomiKo: super fast

struct: got legend on green circle with it

Default avatar.png HenroLST: sheesh

DomiKo: I'm stuck in gold now ;(

jacek: why? i dont forbid you to get into legend

j4at: well C or C++ it all the same for such a simple simulation

j4at: I'm used to extending python using C tho

DomiKo: jacek I beat you once, and now you are stronger! Your NN learned on me

jacek: its different gold boss

struct: yeah its not jacek anymore

DomiKo: really?

struct: its a weaker boss

DomiKo: lol

DomiKo: why?

struct: no idea look at legend size

jacek: at first it was some noob boss

jacek: thats why legend is so big

struct: first it was silver boss

struct: as gold boss

Default avatar.png Aldoggen: they could reset all the ranks

struct: Which made everyone promote

DomiKo: everyone except me...

struct: because you submited your code

DomiKo: ohhh

DomiKo: so they broke it

DomiKo: nice

j4at: I wrote self.angle = radians(degrees(self.angle) + angle)


j4at: instead of self.angle = self.angle + radians(angle)


j4at: I was probably out of my mind when I wrote the simulation KEKW

DomiKo: it's always hard to not make mistakes

DomiKo: it's even harder to debug it

struct: I only have bugs when I read the statement

DomiKo: 143 computing people O.o

DomiKo: I see why, submits are so slow

struct: yeah it was 600 before

jacek: time to promote in csb :imp:

Default avatar.png Aldoggen: hehe :)

BlaiseEbuth: ezpz

j4at: csb ?

DomiKo: csb finally?

DomiKo: mad pod racing j4at

BlaiseEbuth: Cad Sod Bacing

Default avatar.png Aldoggen: cod said bark

j4at: Hehe Coders Strike Back. Well I will it a try, maybe! Last time training my search race NN, time to move on another game. I gave up on passing DomiKo+Recurse. I will pass struct atleast :p

struct: well I could hardcode

j4at: I'm not hardcoding tho :)

struct: didnt you say that you are using the best result your NN found for the validator?

j4at: I'm not even using the validators for training my NN

j4at: well my best NN result is 11000

j4at: but it's not deteminstic so Iso i use best games

j4at: which barely improves it

j4at: to 10800

struct: ah so you sim?

struct: or what do you mean by best games

j4at: Well Its kinda brute force. I have 2 actors after training so i mix their actions

j4at: So its kinda hardcoded but it doesn't make any difference

Illedan: 2 actors?

j4at: 2 policy NNs

Illedan: 2 different NNs? Or 2 outputs?

j4at: 2 different NNS

Illedan: hmm

Illedan: Why 2?

j4at: Its to stabilize ddpg.

DomiKo: you are using Actor Critic methode right?

j4at: ddpg

j4at: with some changes

j4at: yeah ddpg have an Actor + Critic

DomiKo: yeah I know

j4at: So when an Actor learns from a Critic that learns from an Actor. The actor can trick the critic creating a lot of noise that's why I have 2 actors and 2 critics

Illedan: ok, thx

j4at: What are you using DomiKo?

DomiKo: in search race?

j4at: yeah

DomiKo: that should be Double Dueling DQN

DomiKo: with prioritised memory replay buffer

DomiKo: I was training on random maps

j4at: discrete actions ?

DomiKo: yes

j4at: damn

j4at: what actions are using ?

DomiKo: let me check

DomiKo: [-18, -9, 0, 9, 18] x [0, 200]

j4at: bruh

j4at: you are the 13th with that :3

DomiKo: I never really tried doing it. It was a simple test before CSB

DomiKo: it that good or bad?

struct: terrible domiko

DomiKo: :sob:

struct: :)

DomiKo: * domiko crying in the corner*

j4at: well you are doing better than me. So, I can't judge you :p But yeah you should have used continuous actions

DomiKo: not in CSB

DomiKo: that's why I'm not using them here

j4at: nice

j4at: I didn't use a prioritised memory replay buffer because I didn't feel like it

DomiKo: it did a lot for me in CSB

struct: too many NN terminology for me

DomiKo: should we stop?

struct: no lol

DomiKo: from the other game. I've finally used MSmits endgame solver in Oware!

struct: for training?

DomiKo: for validation only

struct: is this the arena version?

DomiKo: I've never really known that scores for positions with only different turn could vary that much

DomiKo: firstly I used scores for position with 150 turns left and my NN was solving like 70% of them

DomiKo: but it couldn't do better

DomiKo: then I looked at score of position with 150, 100, 50, 20, 10 turns left

DomiKo: and I was really shocked

DomiKo: so now I'm training the first agent that uses number of turns left :D

DomiKo: hope it will be better

struct: so its an input?

DomiKo: yes

DomiKo: right now after 20min I can solve 78%

struct: o.o

Illedan: Where can I find that endgame solver?

struct: https://www.codingame.com/playgrounds/58572

DomiKo: from it you can create many samples

DomiKo: it should be close to 200K * 100

DomiKo: I knew that number of turns left was important, but it turns out that it's very important for boards with small amount of seeds

Default avatar.png Secart: 1

DomiKo: 81%!!!

struct: are you using 9 seed or more domiko?

DomiKo: only 9

DomiKo: every position with 9 or less ofc

struct: yeah

j4at: Does anyone have csb simulation ?

struct: the referee?

j4at: yeah

struct: https://github.com/SpiritusSancti5/codinGame/blob/master/Referees/Coders%20Strike%20Back/Referee.java

struct: I think it's supposed to be this

Default avatar.png Aldoggen: it's literally in the description j4at

j4at: I can't use that tho

Default avatar.png Aldoggen: oh

j4at: I need to change it and recompile

j4at: I don't know java

Default avatar.png Aldoggen: what's wrong with it?

j4at: And I don't want to use brutaltester

struct: https://github.com/robostac/coders-strike-back-referee

struct: oh well

struct: cant help you then

struct: I dont have any referee in py

Default avatar.png Aldoggen: do you need to change it or do you need it in python?

j4at: ah that will do thank you :)

Darleanow: hello guys

Darleanow: can, you remind me about the euler's website

Default avatar.png HenroLST: does anyone know a good puzzle to practice simulations?

Darleanow: i forgot the url :/

struct: https://eulerschezahl.herokuapp.com/codingame/puzzles/

struct: HenroLST why not a multi?

struct: or a optimization

Default avatar.png HenroLST: uh not sure

Darleanow: thanks struct ;à

Darleanow: :)*

struct: np

struct: I think search race is nice to learn how to sim

struct: random search can work

Default avatar.png HenroLST: ok, thank you!

struct: but might be hard to get into at the start

Default avatar.png Aldoggen: it's a bit like mad pods lol

struct: well its single player

Default avatar.png Aldoggen: if I see this right

struct: yes

struct: the physics are the same

struct: physics are here in code if you need

struct: http://files.magusgeek.com/csb/csb_en.html

Default avatar.png Aldoggen: my mad pod code would work well for this

Default avatar.png Aldoggen: it's more optimized for single player

j4at: There is a big gap between top 4 and the others in mad pod racing.

struct: NNs vs non NNs

Default avatar.png Aldoggen: lol

Default avatar.png HenroLST: smh

Default avatar.png Aldoggen: I can imagine

j4at: I wonder if i can make it to gold league with my search race code unchanged

DomiKo: you should

struct: even probably legend

struct: you dont need collisions to get legend

BlaiseEbuth: 2 pods

struct: pod collisions I mean

Default avatar.png Aldoggen: my mad pod code does not work in the slightes

Default avatar.png Aldoggen: t

Default avatar.png Aldoggen: yeah, maybe I should have checked input order

j4at: struct Did you try beam search with only [0, 200] thrusts and [-18,-9,0,9,18] angles?

struct: my bot only has 5 moves

j4at: ow

struct: -18 0 18 200 -18 18 0

struct: the replay that I shared used 7 though

struct: -9 9 200 was included

SubZeRoy9: Hello

Jiry_XD: Why does clash of code have bots?

Andriamanitra: this article explains it https://www.codingame.com/blog/clash-of-code-time-has-come-for-clash/

gokubill: halo

gokubill: tôi yêu em

gokubill: i love you

gokubill: Je vous aime

HoangH7: Tau cũng yêu mi.

gokubill: I'm not gay =]]