Chat:World/2021-01-10

From CG community
Jump to navigation Jump to search

ash_rick: https://www.codingame.com/clashofcode/clash/15382962af7ab78df83108b0df8d29d639906f9

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: f

Atomic_Nuclear_WARHEAD: https://www.codingame.com/clashofcode/clash/15383611db6cd3ab4eae4a184b77fca8efef3aa

Default avatar.png ritw777: hey all I'm new here! Just solved the initial if-else problem. really fun IDE


jacek: good for you

Atomic_Nuclear_WARHEAD: hey all I'm new here! Just solved the initial if-else problem. really fun IDE


jacek: :thinking:

Westicles: Some sort of Jungian synchronicity going on here

struct: hello

proogrmr: :rolling_eyes:

proogrmr: hey

Default avatar.png normalkid: bruh

Default avatar.png normalkid: i am so trash

Megha.Akash: https://www.codingame.com/contribute/view/56529ab24fb977bbb361cec3ce8701034586

What does rate mean in this puzzle?

Default avatar.png normalkid: lol i dont think i am gonna no

Default avatar.png normalkid: im in year 7

Default avatar.png normalkid: but i can still do this

Westicles: I think he meant ratio

Atomic_Nuclear_WARHEAD: f

proogrmr: u sus

proogrmr: red sus

Default avatar.png normalkid: yeah

Atomic_Nuclear_WARHEAD: hi

Default avatar.png normalkid: hi

Atomic_Nuclear_WARHEAD: https://www.codingame.com/clashofcode/clash/15384016c23ac06e3afb600d7a951c6d07a09b4

Atomic_Nuclear_WARHEAD: http://chat.codingame.com/pastebin/3637826c-e79f-448f-b182-2743ae2b8db7

Atomic_Nuclear_WARHEAD: https://www.codingame.com/clashofcode/clash/15384016c23ac06e3afb600d7a951c6d07a09b4

struct: stop spamming

eulerscheZahl: stop it

Atomic_Nuclear_WARHEAD: f

eulerscheZahl: oh, hi struct

struct: hi euler

eulerscheZahl: also stop that F spam

Socksfor2: F IN THE CHAT FOR U

jacek: press F

Socksfor2: f

Socksfor2: f

proogrmr: hey

Socksfor2: spam f

proogrmr: stop

Socksfor2: s

Socksfor2: f

Socksfor2: f

Socksfor2: f

Socksfor2: f

eulerscheZahl: my kick doesn't work

eulerscheZahl: struct does yours?

struct: no

jacek: oO

Socksfor2: f

eulerscheZahl: ping reCurse

jacek: UTTT is best game on CG

jacek: done

Socksfor2: watch youtube and ill stop

struct: He changed nam euler

struct: over his profile

proogrmr: yeah he changed

eulerscheZahl: ah, i just realized that too

eulerscheZahl: tried in external client and couldn't find the user

struct: Kick doesnt return an error of why it didnt work

struct: but ban did

struct: "Atomic_Nuclear_WARHEAD was not found on the chat."

proogrmr: it is socks for2

proogrmr: now the username

informathemusic: I think he has an alt

struct: thanks proogrmr, he was kicked

informathemusic: ..oh nvm

proogrmr: :)

proogrmr: yay

informathemusic: :)

Default avatar.png Angecide: just reached top 4 gold in uttt, it feels like legend is just around the corner, but improving the algo gets exponentially harder

jacek: :+1:

struct: Angecide try teccles heuristic

struct: Force player to the same board that you just played

Default avatar.png Angecide: ohh, I did notice that strategy when watching some of karliso's replay

jacek: my bot plays that on his own if you give him more time. seems this is optimal strategy

Default avatar.png Angecide: kinda crazy realizing that top 1 legend has more points in legend than I have in gold

Default avatar.png Angecide: makes you think what kind of algorithm he is using

jacek: ranking is relative

jacek: not comparable between leagues

Default avatar.png Angecide: ah so it is not like elo?

jacek: only within league

jacek: being gold, you probably have more points than most people in legend

eulerscheZahl: and it's not ELO but trueskill

eulerscheZahl: ELO only supports 2 player matches, CodinGame allows up to 8 players per match. So ELO is not a valid option

Default avatar.png ChampionCoder: Hey in the clash result page, there is always something saying "No criteria" What is supposed to go in there?

Default avatar.png ChampionCoder: I mean, it always says "No criteria"

Default avatar.png ChampionCoder: So what should go in there if there is an criteria?

Default avatar.png ChampionCoder: And what is the criteria?

Default avatar.png ChampionCoder: Hello?

dev-tommy: Hello

dev-tommy: how i can check code of my solved puzzles ?

struct: go to the difficulty tab

struct: and then scroll down

struct: https://www.codingame.com/training/easy

struct: if you scroll down it should show the completed ones

Default avatar.png ChampionCoder: Go to the puzzle page, click on "Solve it" and then see the puzzle code

Default avatar.png ChampionCoder: struct I think he's asking the code not the puzzles solved

struct: ah sorry

Default avatar.png ChampionCoder: And can you answer my question?

Default avatar.png ChampionCoder: If it's visible

struct: You can click on results -> history

struct: Is that the form to give feedback on the clash?

Default avatar.png ChampionCoder: Oh never mind sorry it just says the length of code if the type is shortest

Default avatar.png ChampionCoder: Else it says nothing

Default avatar.png ChampionCoder: Thanks anyways

philipplenk:

Default avatar.png ChampionCoder:

Default avatar.png ChampionCoder: Oops sorry

dev-tommy: ok thx i found !

Default avatar.png ChampionCoder: :+1:

dev-tommy: i ask becouse i finished "Temperatures" Puzzle, but now i must finish it again

dev-tommy: i lost results

dev-tommy: oh ok it was the same puzzle but "Code Golfing":wink:

Default avatar.png ChampionCoder: Yeah that is a little confusing

Default avatar.png ChampionCoder: Same puzzles are in both code golf and normal

dev-tommy: ok now i understand ;)

dev-tommy: and its funny becouse my code in code golfing is accepted but in normal puzzle its 90% score ;)

ikadimi: hey everyone i need to create a school planing for students and after some search found the use of Monte Carlo optimization algorithm to score the best possible plannings is there any other way to achieve this or is this the right path?

jacek: could genetic algorithm work here as well? :thinking:

MeSeet: I was wondering the same jacek

eulerscheZahl: or simulated annealing... lost of possibilities

eulerscheZahl: lots*

ikadimi: ok which is best when you have limitations like the number of classrooms and professors and rest periods

eulerscheZahl: i know that my old university used a genetic algorithm for this. there might be better ways but someone got that working at least

eulerscheZahl: they reported an increasing runtime to find a satisfying solution, when you are close to maximum capacity (no spare rooms)

ikadimi: great thank you ima start by doing the Monte Carlo puzzle by aCat and learn how to apply it in this scenario

Astrobytes: isn't that an MCTS puzzle?

ikadimi: yes

eulerscheZahl: independent from your final approach, collect all the inputs first. class room capacities (large lecture won't fit in small room), overlapping lectures, profs wanting a day off, ...

eulerscheZahl: MCTS is not MC

Astrobytes: ^

ikadimi: :)

ikadimi: yes thats why i asked kinda just starting this task

ikadimi: ok then not that

Astrobytes: MC: Generate random solutions and score them. Keeping the best one each time.

ikadimi: any puzzles similar to this problem ?

eulerscheZahl: you can try it in the multiplayer section. i can't think of a puzzle

eulerscheZahl: or maybe something like https://www.codingame.com/multiplayer/optimization/samegame

Astrobytes: Or just try it out on your task. It really is as simple as it sounds.

eulerscheZahl: you won't get top scores with MC but not complete garbage either

eulerscheZahl: time tables aren't that trivial i think. many constraints to consider

eulerscheZahl: just placing random lectures in random spots (pure MC style) will cause a lot of collisions

ikadimi: this one is just for the sports planing so i guess much easier that the hole uni

Astrobytes: No, I meant MC is simple to implement. Not sure it'll fare well for the task tbh.

eulerscheZahl: ah, that makes it much easier :)

ikadimi: the only conditions really is rest days

ikadimi: and space / professor availblity

Astrobytes: Maybe MC will fare better then :)

ikadimi: ok then MC with babysteps it is thanks everyone

eulerscheZahl: and if you've set up the boilerplate code, switching from MC to GA isn't a big step anymore

Astrobytes: ^ You can ask here if you get stuck again, someone's usually around to help out.

eulerscheZahl: my holiday is over, back to work tomorrow :(

Astrobytes: Still working from home?

ikadimi: its a guaranty ill be back :)

eulerscheZahl: yes

eulerscheZahl: will stay that way for a few more months I think

Astrobytes: You guys have a full lockdown or limited?

eulerscheZahl: important stores (groceries) are open, rest is closed

eulerscheZahl: you may go 15km away from your home and meet 1 person from another household

darkhorse64: We have a curfew and remote working is recommended

Astrobytes: Ah ok. And yeah, minimum of a few more months. Probably years over here :P

eulerscheZahl: i'm still an optimist :P

Astrobytes: Pessimism is underrated. I am seldom disappointed ;)

eulerscheZahl: maybe one day we'll realize that the virus is mutating and vaccines aren't that effective after all so we just accept defeat?

eulerscheZahl: like we call in sick for a week with a regular flu

eulerscheZahl: future will show

Astrobytes: It may end up like that, we just have to wait and see, try and control it as best as we can.

eulerscheZahl: with a not even that small group of people seeing the mask as a limitation of their human rights, that's just a firm wish

**BlaiseEbuth is gonna win his Plage.Inc game. :imp:

Astrobytes: Indeed.

Astrobytes: lol, every day is like Plague inc.

eulerscheZahl: plaque? see your dentist

Astrobytes: See your optician!

BlaiseEbuth: Discord on the web chat !

eulerscheZahl: reacting to the typo of blaise

Astrobytes: Oh wait, BlaiseEbuth is playing a beach simulator :P

jacek: another missed opportunity?

BlaiseEbuth: There's hagiss and kouign-amann on my beach Astrobytes, come on !

Astrobytes: :D

Default avatar.png Angecide: top 1 gold, this might be it

jacek: :tada:

Default avatar.png Angecide: damn i might be slightly too weak to be promoted

Default avatar.png Angecide: but just good enough to be first in gold

jacek: spam submit then

Default avatar.png Angecide: I am not entirely sure if that would help

eulerscheZahl: i still don't understand the popularity of UTTT compared to other games

eulerscheZahl: i'm curious, how did you get to UTTT?

eulerscheZahl: k4ng0u just wrote a detailed "post mortem" for Bender4, thanks for the effort :)

MSmits: uttt is popular because tic tac toe is well known and it has a nice wood league

MSmits: then the higher leagues are also relatively easy to sim

MSmits: also it having a high number of players gives a better sense of progression

eulerscheZahl: still you have to find the way from big wide web to that single game on CG out of 50

MSmits: I am not sure how CG suggests games

MSmits: maybe that has something to do with it

eulerscheZahl: even previous contests like codebusters or hypersonic have less players in the arena than UTTT

eulerscheZahl: or at least that's how it was when i checked last

MSmits: also, the talk on worldchat pushes people to try it as well

booleancub3: do you know a good competitive programming site for pract icing

eulerscheZahl: codeforces.com

Default avatar.png Angecide: I wanted to try my minimax algo on normal ttt and then when I saw uttt I got hooked onto it as well

eulerscheZahl: so the really simple wood league helped here, interesting

eulerscheZahl: i try to learn for possible future contributions

BlaiseEbuth: A 3d FPS with tea pots !

eulerscheZahl: can the teapots shoot smaller teapots?

BlaiseEbuth: Sure. Kitchen's law

jacek: actually i was looking for some leaderboard of UTTT, thats how i found codingame

jacek: Angecide how many sims per second you can do?

Default avatar.png Angecide: a ton, i tried testing it on codingames by just doing sims only for 110ms and counting it and it said ~100k sims

Default avatar.png Angecide: but I can only do about ~28k iterations of mcts on turn 2

Default avatar.png Angecide: so that probably has to be optimized a bit more

MSmits: yeah 100k is realistic

MSmits: depending on how heavy your rollout is with heuristics and such

MSmits: I mean 100k iterations of mcts btw, not sims

darkhorse64: That's enough for legend. A more difficult one: what is the winning rate of move 4 4 ? This is to assess if your mcts is correct

MSmits: well it gives a good indication

MSmits: there's more things that can be wrong with your mcts

darkhorse64: A lot !

MSmits: yeah :(

Default avatar.png Angecide: yea, I feel like there might be a bug in my code or I am just not doing enough iterations, but it is incredibly hard to debug

MSmits: though this weekend i worked on a negamax with ab and TT and I must say I made a lot of booboo's as well

MSmits: Angecide yes it is

darkhorse64: It's also easy to have something that works ok as p1 and completely bogus as p2

reCurse: It's hard to lose as p1

reCurse: You have to work for it

Default avatar.png Angecide: would you say that in theory with perfect information that p1 can force a win?

jacek: MSmits do you have history of your meta mcts? i wonder how the graph of the moves oscillate

reCurse: You need to solve a game to prove it

Default avatar.png Angecide: yea I just meant your gut feeling

reCurse: So in the absence of it, it is overwhelmingly in the direction of p1 always wins

MSmits: jacek i dont have a history, move move 0 was near 50% for a long time, then moved up to 55% and then back down to 47% at a minimum. Going back up again

reCurse: I suspect it's the board count that makes it all messed up

reCurse: But didn't investigate

jacek: no prove, but based on many games, how mcts favor the 1p, it seems it is likely uttt is win for 1p

jacek: especially with board counting

reCurse: Stupid rule imo

jacek: they added it to avoid draws?

reCurse: lol...

reCurse: That causes *more* draws

MSmits: jacek i am very near certain uttt is solved as win for p1

MSmits: whenever i see a p2 win

MSmits: i can trace it back to a mistake of p1 earlier

MSmits: thats how i counterbooked players that used books as p2 (bad idea)

MSmits: if you book, only do it as p1

jacek: i wonder when the cheese will be out

reCurse: Are you still booking in the arena

MSmits: hmm, i think it only counters yours as p2 currently, no other hardcoding.

MSmits: unless you took down that book

MSmits: or maybe you didnt have one and its just really deterministic

jacek: interesting https://wikidiff.com/chess/cheese

reCurse: Wonder if I should be flattered or upset

MSmits: dont worry about it.... ill be redoing my entire bot when i have the time

MSmits: i have locally hardcoded karlisos bot as p2

jacek: being deterministic is good if its going the only good moves available

MSmits: can get 100% winrate vs it, but i didnt even submit

MSmits: it's boring

MSmits: he doesnt even have a book

reCurse: So with me it's fine but with him it's boring? Meh

MSmits: its just limited variation of play

MSmits: nah i did this a year ago recurse

MSmits: for your bot i mean

jacek: in 2020? so old

MSmits: also, i can take it off and have like a 53% overall winrate vs you

reCurse: Nah don't change anything

MSmits: with it, it is 100% against 35% or something as p1, and p2, without it, it is 70% vs 35%

jacek: and you finally tried negamax. what for

MSmits: so it just changes the 70% to 100% as p1

MSmits: the book i man

MSmits: mean

MSmits: jacek for study project

MSmits: yavalath

reCurse: Should be 70% in my favor now

MSmits: oh you submitted a new bot reCurse?

reCurse: No

MSmits: whats different then?

reCurse: Well new bot yes, submit no

MSmits: ohh ok

MSmits: cool

MSmits: does it beat karliso?>

MSmits: cuz that one is ridiculously strong

reCurse: I think I finally found one yes

MSmits: oh did you do machine learning?

reCurse: Yes

MSmits: ahh ok

jacek: :O

MSmits: I'm sure it can still be counterbooked when it is p2, but as i said, not going to

Default avatar.png Angecide: how can you tell when someone books vs no books? is it if they move deterministically?

reCurse: No books

reCurse: But maybe deterministic

MSmits: yeah

reCurse: Then again I'm even more dependent on the CPU I get than before

reCurse: It's frustrating but I asked for it

MSmits: thats the problem, good bots dont have that much variation in play

reCurse: Looks like 70% vs him so far

MSmits: interesting

eulerscheZahl: your fights at those games reminds me of goats banging their heads against each other

reCurse: Well tbh

reCurse: I really didn't care about uttt at all

reCurse: Just needed an easy target for my new ML/RL pipeline

reCurse: And maybe give me some peace in the process

MSmits: so you did care!

reCurse: ?

MSmits: on some level :P

reCurse: No

MSmits: why would it give you peace

MSmits: you had inner turmoil from uttt

reCurse: That it wasn't all for nothing

MSmits: ohh ok

jacek: you like UTTT dont you, reCurse https://i.kym-cdn.com/entries/icons/facebook/000/009/086/SMILE.jpg

reCurse: No I just lacked an easier target

MSmits: isnt oware just as easy?

MSmits: or othello?

reCurse: Oware is community

MSmits: ahh

reCurse: I'd rather not touch those if I can help i t

jacek: but you touched oware

Astrobytes: Oware touched him first.

reCurse: Yeah and it made me discover the joys of limited submits

MSmits: ah

jacek: and stupid robo is 1st :(

MSmits: if your bot is much stronger than others, it doesn't matter that much

Astrobytes: Can't see that situation ever improving tbh

reCurse: Of course it matters

reCurse: My CSB bot when I resubmitted it

reCurse: Has >80% winrate against everyone

reCurse: Barely made it 1st at 90% submit

reCurse: So imagine on a community

reCurse: It's dumb

MSmits: it's a large league though

MSmits: i will argue that the imbalancedness of uttt + better submit is still worse than a more balanced game with limited submit

MSmits: uttt is pretty bad like that

MSmits: even with swap

MSmits: all those draws...

reCurse: That's a defendable position

reCurse: Outweighted by my sheer hatred for it though

MSmits: haha ok

reCurse: I don't understand why it's a random iteration that's crushing it

reCurse: It's 25 generations ago

reCurse: But it's doing it

reCurse: ML is hard...

MSmits: as i said, karlisos p2 play is deterministic enough that you can counter book it with less than 100 moves or so.

MSmits: that would mean some iteration of your NN might do this as well

reCurse: I believe it generalized well enough though

reCurse: It's consistent against top3

reCurse: At least

MSmits: sure, but karliso is wayyy above me and tomatoes

reCurse: But very fragile, an iteration later and it's average

MSmits: in strength

jrke: which multi?

MSmits: i lose 100% as p2 vs him

eulerscheZahl: uttt

jacek: you have NN for uttt?

reCurse: Yeah

jacek: :+1:

MSmits: reCurse do you have a random feature?

reCurse: ?

MSmits: like, add a random float to the score of each move

MSmits: and then pick one

reCurse: I'd rather not

MSmits: it might help with the difference between those iterations

MSmits: otherwise this really good bot could collapse if karliso adds or removes one heuristic from his bot

reCurse: Oh I have a lot of things for training

reCurse: But for evaluating why would I do that

reCurse: Well I'm assuming that if it maintains a good winrate vs the top 10

reCurse: Then it's generalized

reCurse: I don't care about overfitting with books it's dumb

MSmits: its not just books, its also deterministic bots playing nearly perfectly

reCurse: You can probably counterbook my new one

reCurse: It's an escalation war I have zero interest in

MSmits: which means one small change and a 100% winrate becomes 50% or less

MSmits: reCurse counterbooking only works as p1 really

MSmits: so at best it would be 50%

Default avatar.png Angecide: btw, I was wondering with all these books, how do you import it into your codes? is it just a large piece of hardcoded tree containing the actions?

MSmits: as your bot will beat my p2 bot 100% and i would beat yours

reCurse: I will take the random CPU attribution as random float

MSmits: that helps to some extent

reCurse: Because performance difference is extreme

reCurse: Not even consistent in a series I think

reCurse: Which is stupid

reCurse: Angecide: It's just serialization embedded as a constant

reCurse: As for my complaints about GPU, I estimate my training would be 5x faster with rtx3080

reCurse: It's crazy

MSmits: you train on gpu?

reCurse: Yeah

reCurse: I went all out

MSmits: did you use any tools now?

MSmits: you didnt create i mena

MSmits: i know at first you made everything yourself

reCurse: Yeah I'm leveraging libs now

reCurse: I've gotten more confident in my understanding

reCurse: Besides gpu programming is a can of worms of its own

jacek: "What I cannot create, I do not understand"

MSmits: yeah

reCurse: Is that the wisest thing you've said jacek

MSmits: i am amazed that people get anything done on a gpu besides the graphics processes it was intended for

jacek: i know more random quotes

MSmits: my short foray into compute shaders was painful

jacek: but simply i dont understand NN i didnt create from scratch myself

reCurse: cuda is very good

reCurse: But it's a very different mindset

eulerscheZahl: i looked into cuda long ago but gave up quickly :(

reCurse: For me it's enough to have a good intuition of how things work by doing it from scratch

reCurse: Then you can better leverage libraries

eulerscheZahl: in the end it's always the question what i'd like to do first in my spare time. sadly my motivation to start such a complex thing dropped a bit since i'm writing code in my job as well

eulerscheZahl: i miss my student days

reCurse: Sure, makes sense

reCurse: I see those as completely different

eulerscheZahl: but both are mentally challenging and I want some time to rest as well

reCurse: Job often doesn't align with curiosity and learning interests

reCurse: Though I'm working on a ML project at work now, and that's thanks to practicing it with CG as motivation

reCurse: Wouldn't have gotten into it or thought it would be for me otherwise

reCurse: Do servers still have hiccups?

Salted: Thanks to CG I also got into ML. My coders strike back bot is the product of genetic evolution with a neural network. I've even been lucky enough to make some side cash doing ML now.

reCurse: You used a GA to optimize a neural network?

Salted: Yeah, to tweak the weights of the network. They compete in a league against the "best" bot and the winner becomes the best bot to beat. I used the brutal testing stuff.

reCurse: Interesting. Why not gradient descent?

reCurse: I never got the point of neuroevolution

Salted: I wasn't sure how to use gradient descent when I didn't have a scoring function. Just a single point indicating that that it won/lost a match.

reCurse: That's a scoring function

MSmits: temporal difference learning is supposed to help with this problem Salted

MSmits: i havent done this myself, but i know its an answer to this

reCurse: Kind of orthogonal to optimization though

MSmits: no idea, i will trust you on that

reCurse: You can do TD with gradient descent

reCurse: Don't, I'm no trained scholar

MSmits: ok :)

reCurse: Just the product of google searches

MSmits: same

Salted: I'm no trained scholar either.

MSmits: though, i will say that on dots and boxes i am almost a trained scholar :P

MSmits: never read so many papers

reCurse: There are "so many papers" on D&B?

MSmits: nothing else though, not even mcts, i half read a few is all

reCurse: I've been reading papers nonstop since I joined CG

MSmits: reCurse maybe a dozen or so, i meant it proportionally

MSmits: so i know most of the body of knowledge on this

reCurse: Ah ok

MSmits: mikla still beat me though, when i have time i will get back to it

Salted: See I thought gradient descent would only work if you had control of the data. Since it invoves another player then changing the network affects how the data of the input too and not just the output which is why I couldn't use it.

reCurse: Oh sure it's very noisy

reCurse: Doesn't mean you can't optimize it

Salted: oh, sorry, that was terrible. The input data gets changed when changing the network. So gradient descent doesn't work

reCurse: RL is the far west of ML

Salted: Oh, but I can watch another player and create a network that mimics their play style

Salted: That's easy

reCurse: Sure that's the typical supervised learning

jrke: how do you store weights of training?

Salted: I did try the traditional supervised learning with the game data in the beginning but it never made bots that worked well

reCurse: Same way you store constants

reCurse: Except with more work

Salted: I store them in a compressed pickled array

reCurse: And no 100k chars is not an issue

Salted: It is an issue for 100k. So I can store maybe 17,000 floats

Salted: Well, it's not a normal issue

reCurse: You'll just have to take my word for it

jrke: how in 17000 floats only can you give me some reference link for it

Salted: I've tried running more and the script won't run.

reCurse: Not giving away all the secrets either

jacek: it can be more than 17000 floats. you dont need that resolution

jacek: i.e. for bt i have 49152 weights with about 'int16' resolution

reCurse: bt?

jacek: each weight using 1 character

jacek: breakthrough

reCurse: Oh

Salted: Oups An error occurred (#314): "Submitted code is too big"

Salted: That's what happens when I go over 100k

reCurse: Yeah

reCurse: 100k characters

jacek: you need to be more creative in those 100k

reCurse: It's like those 4kb demo competitions

reCurse: You have to squeeze

Salted: This is what my network looks like for example, but another 80k of this

Salted: eNqEmnk018Hb9+2y71ERiRCVLet3rg8qIsq

jacek: and this is mine 憄晁攱攰斨敏旁斉斡斝斒旭抻桥斖旟斥敒斣旇敫既斎旴惤揓敷旭斗旰架敱攻旊斬旬搞斨旦旬敳斖旃旳敼旔旈攴愍...

SwagColoredKitteh: oh hi

Salted: Oh man

Salted: lol

SwagColoredKitteh: hope you're all having a great day

struct: hi

SwagColoredKitteh: hi struct

Default avatar.png PIRT: hi

SwagColoredKitteh: hi PIRT

jacek: all glory to unicode

SwagColoredKitteh: and this is the story of how i became a greeter bot

eulerscheZahl: because you can't hug people anymore

SwagColoredKitteh: :'(

SwagColoredKitteh: been messing around with crystal rush, pretty fun game

reCurse: I wonder when a hug will be considered a natural thing again

reCurse: Now I'm depressed

SwagColoredKitteh: join the club

SwagColoredKitteh: at least i could channel it into writing code; i call it depression-driven development

MSmits: :hugging:

SwagColoredKitteh: i think hugs will quickly become a natural thing again between people who know one-another

SwagColoredKitteh: but it may be very slow for strangers

reCurse: Not even sure about the former

SwagColoredKitteh: hey MSmits :hugging:

reCurse: The virus stigma will last for a long time I think

reCurse: At least for reasonable people

MSmits: hey :) and bye getting some food

SwagColoredKitteh: enjoy MSmits!

eulerscheZahl: hey bye to you too

Salted: Hugs are nice. I can't wait for basic eye contact to come back. It feels like people are really being cold these days.

SwagColoredKitteh: reCurse well, yeah, it still only will resume after we've all been vaccinated

SwagColoredKitteh: i'd assume at least

reCurse: That's optimistic

SwagColoredKitteh: optimism is all i have right now

SwagColoredKitteh: there's also optimism in the assumption that everyone will get vaccinated

SwagColoredKitteh: .. a lot of optimism

reCurse: Or that it will re-enable normal life

reCurse: My optimism points to 5 years now

Astrobytes: I have none. We're doomed I tell thee :P

SwagColoredKitteh: maybe 5 years for everything to be 100% normal again?

reCurse: Let's say 90% normal

reCurse: But yeah

SwagColoredKitteh: but i think there will be a slow transition

SwagColoredKitteh: i'd already be very happy if i could meet with & hug closer friends

reCurse: True

SwagColoredKitteh: the rest i can live with missing for a far longer time

Astrobytes: My grandfather was put in a residential care home towards end of summer, I haven't given him a hug since March.

reCurse: :(

Salted: ok, for this gradient descent concept. Lets say each game is 200 turns long. That means there's been 200 inputs and outputs. But only 1 label to represent them all. That's like labeling 200 pictures cat/dog but only knowing a majority of the guesses were right. But there's no way to know which were good and bad. How might I go about using gradient descent?

SwagColoredKitteh: :(

reCurse: With a LOT of games

SwagColoredKitteh: yeah, a lot of kinds of ML need a ridiculous amount of time and data to converge

reCurse: My strongest uttt sits at 2M games right now.

Salted: The issue I have is that it may only ever need to do better at 51% of the games to end up always winning. So it doesn't converge to a very nice bot.

reCurse: ?

Salted: Maybe I'll read more about it and see if I can get gradient descent to work. Have you implemented a nn bot from gameplay before?

reCurse: Sure

reCurse: CSB

SwagColoredKitteh: ah yes CSB, the one true game

Salted: I really wish more games were similar in style.

Astrobytes: there's Search Race in optim

reCurse: Poor fit for nn

eulerscheZahl: mean max bitrunner 2048

Astrobytes: Oh yeah, BR

eulerscheZahl: and fantastic bits, poker chip race if you extend the "similar" term a bit

SwagColoredKitteh: search race was fun but to get any better i need to refactor so much code

Astrobytes: lol, didn't Illedan create SR so he could use it for developing an NN? :)

SwagColoredKitteh: i havent done fantastic bits in so long

Salted: Oh, I haven't even joined bitrunner before. Nice

Astrobytes: *a NN

reCurse: Ok, my bad

reCurse: It's great to get something started

reCurse: It's bad fit to compete with

Astrobytes: Ah, gotcha.

jacek: an* NN. i always pronounce NN as ehn-ehn

reCurse: So you say an URL? :thinking:

reCurse: Nonsense

jacek: yu er el, so no

reCurse: :sweat:

Astrobytes: When in written form I tend to expand abbreviations mentally as I read. Hence a NN. In speech, of course if I used the NN abbreviation I would say 'an NN'.

jacek: AutomatonNN how should we call you?

AutomatonNN: why do you get out of wood 3 ?

BlaiseEbuth: He's right. Get back to wood 3 jacek.

eulerscheZahl: regarding a/an <acronym> I agree with jacek

Astrobytes: Both work in written form. In speech they don't.

Astrobytes: btw, eulerscheZahl (or anyone): the 'recolored' validators in samegame are recolored how?

eulerscheZahl: colors swapped,shiften/whatever

eulerscheZahl: didn't bother, just made a color-independent hashing

eulerscheZahl: first (at 0,0) is always color=0

eulerscheZahl: next value I find gets a 1 and so on

Astrobytes: Yup, makes sense. What the hell was the point in the recolored tests?

Astrobytes: *of the

eulerscheZahl: someone didn't believe us that we would solve it online anyways

eulerscheZahl: or expected his own students to do it online and wanted a more stable ranking for those

Astrobytes: Eh. Well...

eulerscheZahl: i argued against it. that's also why I didn't approve

Astrobytes: I remember now, yes.

eulerscheZahl: at least i could convince him to have a round-based game. not giving all commands in the first turn

eulerscheZahl: so you get the next state which helps debugging your sim

eulerscheZahl: + easier to enter

Astrobytes: Yeah, I remember that conversation. I wasn't bothered but it made sense.

SwagColoredKitteh: a bunch of my bots were crashing and i fixed and reuploaded them

SwagColoredKitteh: woo, chores

eulerscheZahl: some of mine died too :(

eulerscheZahl: mono -> .net core

jacek: use D [solved]

SwagColoredKitteh: cg switched to rust 2018 and i was using rust 2015 paths in use statements, and using try! instead of the ? operator

eulerscheZahl: and CG didn't add the System.Drawing.Common

jacek: does CG have java swing?

SwagColoredKitteh: wait, C# drawing?

eulerscheZahl: rust has release mode now

eulerscheZahl: i draw bitmaps for debugging

eulerscheZahl: easier to spot bugs than text output

SwagColoredKitteh: yeah i saw, so i also switched a bunch of my precompiled bots to just having the source code there

SwagColoredKitteh: much nicer

SwagColoredKitteh: always hated the precompilation hack

jacek: how can you show the bitmaps?

eulerscheZahl: i just draw them offline

jacek: :x

eulerscheZahl: and don't call the function online but still have the code

eulerscheZahl: now I had to remove that code for the uploaded version as it won't compile anymore :/

SwagColoredKitteh: what i used to do was to print a compact format to stderr for drawing some things

SwagColoredKitteh: and copypaste that

SwagColoredKitteh: but it's really not a great solution

eulerscheZahl: https://www.codingame.com/playgrounds/53705/contest-tools-and-workflow/introduction my setup

SwagColoredKitteh: would be cool if CG had some way to just directly draw on the game

eulerscheZahl: except I use VS Code now and no more MonoDevelop

SwagColoredKitteh: maybe using some prefix on stderr or sth

Astrobytes: Yeah, that was one cool thing about RAIC. CG version would have to be a little....simpler.

eulerscheZahl: RAIC has such a debugging framework

eulerscheZahl: damn

Astrobytes: :D

SwagColoredKitteh: just being able to plot the next move choices would be such a huge timesaver :P

SwagColoredKitteh: or being able to overlay a grid in crystal rush, as i'm calculating a few for stuff like how far a cell is from the closest radar

eulerscheZahl: hm, even sounds like an interesting project to bring such a debug interface to CG

SwagColoredKitteh: just for that would be nice, but it'd be even better for combinations of such grids, because that can lead to quite unintuitive behavior sometimes

Astrobytes: Considering we now have 3D, I don't think it should be such an awful project :)

eulerscheZahl: but more limited as it would be too easy to kill pixi

SwagColoredKitteh: this score here, combined with that score there, etc

AbacusCdilla: Hi Everyone

jacek: good evening

dev-tommy: Hi i try solve Nintendo Sponsored Contest ;)

jacek: good luck

dev-tommy: XOR is easy to reverse , but AND cant be easy reversed. I must combinate with another bits

MSmits: show those bits what for

dev-tommy: not figured out yet that

MSmits: it was meant to be encouraging

Westicles: I've always wondered what kind of work they do at Nintendo, that they are looking for these kind of skills...

dev-tommy: i try with 4bits numbers first

MSmits: why not start with nintendo 8 bit

dev-tommy: it will be easy to check on paper

Astrobytes: yes, it's a somewhat niche puzzle

dev-tommy: yes it is more of a mathematical puzzle than a programming one

Astrobytes: And quite a specialised mathematical puzzle at that.

dev-tommy: but i like this kind of puzzle

Astrobytes: Sure, go for it! :)

jacek: you like math? :thinking:

Astrobytes: *maths

jacek: s/a/e

dev-tommy: yes

dev-tommy: the decode answer is symmetrical so i think I'll find the missing bits information there

Westicles: I also like rearranging furnitures and eating sushis

LooMATE: Afternoon / Evening all - to a total newb to coding, which language would you suggest I start with - I know its likely a nuanced question but I am happy for personal opinions..!

JonLapensee: is website lagging or it is me ?

Default avatar.png shefoo: Try javascript it's easy and fun, i came out of java and i loved javascript so much

Westicles: Python is easy and useful

Default avatar.png shefoo: @westicles Agree

Default avatar.png 26: xylophone

Default avatar.png 26: start dat shit jk

Ragi: people use code like this in code clash with python3 exec(bytes('浩潰瑲洠瑡㭨㵲㬰㵩㬱...

Ragi: how do you encode to this form

Default avatar.png Jk7292: idk

Default avatar.png 26: start the match derek

Default avatar.png APJames: is this for coders strike back?