Chat:World/2022-05-29
Ajaiy: gm
antiwonto: [auto] Hey Ajaiy, here is a :taco: for loggin in today while it is quiet. You now have 12 tacos
chaosabordine: wat with?
5DN1L: with this probably:
5DN1L: https://how-to-properly-ask-a-question.surge.sh/
chaosabordine: just the url lol
chaosabordine: A: Hi, here is a problem that I have: [the problem] I mean, that's way better, isn't it?
5DN1L: yeah, "help" is too broad
chaosabordine: just checking is it a comment on my earlier questions?
chaosabordine: this -> hello community i have an hist history question sry coding legend history that is. any buffs on that subject?
chaosabordine: actuallu. i thought you also zere the person that posted "I need help" then the link to the url lol
chaosabordine: were
5DN1L: that'll be hilarious or absurd :smirk:
chaosabordine: i went with absurd..
chaosabordine: but found it hilarious.
jacek: :upside_down:
antiwonto: [auto] ':upside_down:' was defined as ' ɥǝllo '
waterproofsodium: ah ha
antiwonto: [auto] Hey waterproofsodium, here is a :taco: for loggin in today while it is quiet. You now have 37 tacos
jacek: :rage
waterproofsodium: the bot recognizes very well that each of my mindful contributions is at least worth one taco!
waterproofsodium: :taco: jacek
antiwonto: [auto] waterproofsodium has awarded jacek 10 tacos. jacek now has 49 taco. waterproofsodium now has 38 taco
waterproofsodium: this is the oldest one
waterproofsodium: I had to get rid of it :p
jacek: hm?
tutubalin: puzzle search is not working
antiwonto: [auto] Hey tutubalin, here is a :taco: for loggin in today while it is quiet. You now have 2 tacos
5DN1L: euler's site is down right now
5DN1L: wait until start of next month
5DN1L: it'll be up again
tutubalin: thanks!
5DN1L: np
tutubalin: how can i find the puzzle from the last event?
tutubalin: i'd like to see who won
5DN1L: i guess you get an email for that?
5DN1L: not sure
tutubalin: for the 3rd phase - no
tutubalin: but yeah, i could find it via previous emails
jacek: Automaton2000 where are my tacos
antiwonto: [auto] Hey jacek, here is a :taco: for loggin in today while it is quiet. You now have 50 tacos
Automaton2000: i tried to use the same code
jacek: phew
DomiKo: easy taco
jacek: was that unicode thing any helpful?
DomiKo: yes it was
DomiKo: trick with normalizing is soo good
StevenV: Hello guys
DomiKo: hello
struct: hi
44AA44: 44AA44
jrke: anybody have idea using what library or tools we can put aur trained RL model weights on CG?
antiwonto: [auto] Hey jrke, here is a :taco: for loggin in today while it is quiet. You now have 2 tacos
Hunter64TheOne2: Hello can someone (please) help me learn c++ and python? Thank you.
noobcoder42: just code and start practice
Hunter64TheOne2: yes, I thought so, but...
jacek: jrke how big is your model?
jacek: if small enough, you can manually copy & paste the weights ;d
jacek: oware battles revived eh
DomiKo: a lot of people submitting I see :D
antiwonto: [auto] Hey XJDream, here is a :taco: for loggin in today while it is quiet. You now have 1 tacos
XJDream: Look at this NFT anyone like?
5DN1L: XJDream Don't paste these links in the chat
XJDream: oh sorry
Wizard-Dude: when will the next competition start?
5DN1L: no news yet
jrke: jacek i am just asking which libraries i can use whose weight can be put on CG
jrke: like stable baselines3?
Uljahn: what do you mean? weights are just arrays of numbers, that's the only way to put your model on CG
Uljahn: there is no special ML/DL/AI support on CG, other than general purpose libraries such as numpy
j4at: We need tensorflow :'(
j4at: or even better pytorch
DomiKo: numpy is really enough
j4at: Converting my model from pytorch to numpy is a pain in the ass.. Because the model is complicated
DomiKo: I'm doing the same and it's like 20 lines of code for me
DomiKo: you only need to take 2D/3D arrays from your array
DomiKo: notthing special
j4at: What about the operations MultiheadAttention, custom PositionalEncoding and other custom layers, i'm using. I need to convert everything to numpy. But well it's understandable why they can't add pytorch/tensorflow.
struct: I dont understand the training part well enough yet
jacek: im not aware of 'easy' way to convert weights from libs to your needs. one plus for doing all the stuff yourself :v
EvilClownVonKirschenwald_825a: Are there any tutorials on how to utilize this site for lessons? I would like to use it for a last-week section for some AP CS Principles students. I did a basic coding challenge in Python but can't figure out how to get to the next step. Thanks.
Stromtropers: hello there please dont banned me I am not gonna do something bad I promis wont do something bad
5DN1L: EvilClownVonKirschenwald_825a:
5DN1L: https://www.codingame.com/blog/1-hour-to-teach-programming-with-codingame-in-fun-way/
5DN1L: https://www.codingame.com/blog/one-hour-learn-bot-programming/
EvilClownVonKirschenwald_825a: Thanks.
Stromtropers: i am going to the site
5DN1L: EvilClownVonKirschenwald_825a, there's one more:
5DN1L: https://code.org/files/coders_strikeback_teacherguide.pdf
MAWAAW: https://www.codingame.com/clashofcode/clash/242984680be05204f468734b7eada2d7d100349
antiwonto: [auto] hey MAWAAW dont paste those links here. Use the channel #clash
jacek: coders strike back? that takes me back...
5DN1L: those were the days
Stromtropers: how do we create a communite contest
jacek: :(
Stromtropers: dos somebody know
5DN1L: What community contest are you thinking of?
Stromtropers: any
5DN1L: That's too vague
5DN1L: Give an example of the contest format
Stromtropers: I want to create one
jacek: oO
5DN1L: If you can't give any details, we can't asnwer you, Stromtropers
Stromtropers: ok
Stromtropers: clash code contest
5DN1L: are you talking about creating a single clash of code puzzle?
Stromtropers: yes
5DN1L: https://www.codingame.com/playgrounds/40701/help-center/creating-your-own-puzzle
Stromtropers: thank
Stromtropers: what can I put like input description and output
5DN1L: Refer to other contributions and puzzles for some examples
Stromtropers: I am contributing a clash of code
5DN1L: If you have no idea, please practise some first before you try to create one yourself
Stromtropers: and where I practise
5DN1L: https://www.codingame.com/multiplayer/clashofcode
Darleanow: Hello 5dn1l :)
5DN1L: Hello Darleanow :)
Stromtropers: you are frend
MAWAAW: #vn
Stromtropers: #hello
Stromtropers: 5DN1L
Stromtropers: are responding
5DN1L: If you have any questions, just ask. It's not always me that will be answering here
Westicles: Welcome Stromtropers, have a taco
Westicles: https://imgur.com/a/AsBCNuF
Stromtropers: have taco what
Stromtropers: 5DN1L why do my computer is making sound with code in game
nicola: Maybe when someone is poking you.
nicola: Like this: Stromtropers.
Stromtropers: HOW DO A LIKE THIS nicola
nicola: What?
Stromtropers: like this :thumbsup:
.fill.: #hello
Stromtropers: #.fill.
Stromtropers: # #
Stromtropers: # e
5DN1L: that's all for you for today
JabaDUDE: Hey guys new to the site. Seems like a fun way to get good but... uh... well you see, and this is awkward, I have no idea wtf I'm doing. Can anyone assist me
5DN1L: Have you learned any programming languages before?
JabaDUDE: I'm learning my first one, Javascript
Razovsky: haaaaa
Razovsky: noooooooooo
5DN1L: You see, this is a practice website rather than a tutorial website
Razovsky: nothing can save you
JabaDUDE: My problem isn't the language, it's how to navigate the site and understanding what i'm looking at
Razovsky: flee
5DN1L: JabaDUDE you can practise by playing puzzles here: https://www.codingame.com/training
5DN1L: and you can play multi-player games here: https://www.codingame.com/multiplayer
5DN1L: More details here: https://www.codingame.com/playgrounds/40701/help-center/im-new-to-codingame
JabaDUDE: Im doing the training and looking at The Descent, I guess I'm not sure what its wanting me to do.
Wizard-Dude: question, how would I calculate the impulse of two objects when they collide, if both objects have a mass of 1 and all I know the velocity of both objects?
jacek: youre probably struggling with input/output stuff
jacek: even more advanced people are struggling at first
5DN1L: JabaDUDE you're supposed to revise/complete the code so that it will solve all the test cases/validators
5DN1L: you read the inputs, process them, and write the answers as outputs
5DN1L: write refers to "console.log" in javascript
jacek: in the hints you even have complete solution. but generally you have, in loop, the list of mountains in inputs and you need to print desired output in each iteration of loop
JabaDUDE: I know that, I guess I'm confused by the text. Am I shooting the last mountain or am I shooting all the mountains?
5DN1L: that is just visualisation
jacek: shoot the highest mountain
5DN1L: what you really have to do is to deal with text inputs and outputs
JabaDUDE: gotcha, so I'm just outputing the last index
5DN1L: outputing the index of the highest mountain
5DN1L: index in the list of heights you are given
struct: Hi YurkovAS I see you are also doing STC, are you using random search?
5DN1L: e.g. if inputs = [7,6,5,4,3,2,1,0]
5DN1L: you output 0
5DN1L: because 7 is index 0 in the inputs
jacek: Wizard-Dude if youre doing simulation for mad pod racing i think there are some implementations already
JabaDUDE: Ok so each turn the 'high mountain' will be a diff index and my code has to locate it thru loop, and 'fire it'
5DN1L: yup
JabaDUDE: Gotcha. I guess I was reading too much into the description. Thank you.
5DN1L: np
jacek: like http://files.magusgeek.com/csb/csb_en.html
frits.becker: http://chat.codingame.com/pastebin/07721f91-f65e-4f83-95be-aedf4aaec0ee
5DN1L: you can post your question on the forum
5DN1L: i'll amend the title of the post now
5DN1L: https://www.codingame.com/forum/t/community-puzzle-guessing-n-cheating/61810
frits.becker: I finally understand what I am doing wrong. thank you very much!
5DN1L: np :)
YurkovAS: struct hi! I'm use mcts
struct: oh nice
struct: Im redoing my unfinished bot
struct: Trying to use bitboards here
YurkovAS: + fast connected components https://www.codingame.com/playgrounds/58038/fast-connected-components-for-6x12-bitboard
struct: didnt use that but I think the floodfill is fast enough
struct: There are plenty of tricks for non bitboards too
sdfk: why dont people use java?
struct: it's verbose
struct: you need to type a lot
struct: :taco: jacek
antiwonto: [auto] struct has awarded jacek 10 tacos. jacek now has 50 taco. struct now has 48 taco
sdfk: and i always lose the shortest battles
PatrickMcGinnisII: :taco: struct
antiwonto: [auto] PatrickMcGinnisII has awarded struct 10 tacos. struct now has 58 taco. PatrickMcGinnisII now has 32 taco
PatrickMcGinnisII: taco horse race
struct: thanks patric
struct: sdfk use ruby to beat python
Westicles: :taco: BlaiseEbuth
struct: heres a few tricks
antiwonto: [auto] Spread the love. Give tacos to someone else, not someone you spam with tacos.
struct: https://github.com/siman-man/ruby-golf-style-guide
sdfk: i only know java from my school
sdfk: python looks pretty simil
jacek: dont bring knife to a gun fight...
5DN1L: bring a white flag?
jacek: huh
jacek: struct :taco:
antiwonto: [auto] jacek has awarded struct 10 tacos. struct now has 68 taco. jacek now has 41 taco
Wontonimo: https://www.youtube.com/watch?v=j-fWDrZSiZs white flag
5DN1L: capture the flag: https://lab.reaal.me/jsrobot/
Wontonimo: waving flag https://www.youtube.com/watch?v=WTJSt4wP2ME
Wontonimo: that's sad DOMINER
Wontonimo: if only there was a way to help you
Westicles: here you go DOMINER
Westicles: https://imgur.com/a/AsBCNuF
jacek: oO
Narcyssus: is it possible to open a clash report by id from channel ?
Wontonimo: gotta go
jacek: you have clash reports in notification
dtator: Is there any way to clash only with persons using the same language?
struct: Yes in privates clashes or during clash events
Narcyssus: jacek, got it, thanks
antiwonto: [auto] That taco command has moved to the #taco channel
-Ace: Hi
CarajoRosa: why the descent take so long to understand xd
jacek: hm?
struct: you just need to understand how puzzles work I guess
ValentinBrault: function find in js doesn't work in codingame?
jacek: array.find()?
ValentinBrault: yes
struct: it seems to work
jacek: http://chat.codingame.com/pastebin/7ceaee40-d0c7-43eb-b38b-0bebcaaaec01
jacek: it works
ValentinBrault: ok maybe a error from me
ValentinBrault: thanks
struct: jacek how many weights does your NNs usually have?
jacek: oware has 81k
jacek: i keep them around 50k-80k
struct: any reason for that?
jacek: around 50k if i want to keep the code still readable. more for performance
jacek: i use one-hot inputs which are not size friendly, but fast
struct: I see
struct: I was thinking if L1 cache would help but it doesnt really fit there
struct: it would need to be like 8k weights
jacek: 8k weights? what is it? an NN for ants?
struct: would still beat my bots
struct: :D
DomiKo: "NN for ants" :joy:
Aligg5: :slight_smile:
Peter_Riedel: (╯°□°)╯︵ ┻━┻
Wontonimo: hi
antiwonto: [auto] Hey Wontonimo, here is a :taco: for loggin in today while it is quiet. You now have 27 tacos
Wontonimo: yeah
struct: hi
Wontonimo: how's it going struct?
struct: good, trying to finish my stc bot
struct: what about you?
struct: are working on something?
Wontonimo: nope. I've been chatting just a tiny bit with jacek and ImSeaFu about NNs
Wontonimo: don't know if I've been a help
Wontonimo: probably I'd be able to help more once I actually use NN on CG
struct: you should try it
struct: I would help if I had a clue how to do one
struct: lol
Wontonimo: yeah
Wontonimo: :grin: thanks
struct: I could help writing the past forward stuff in c++ in avx or something
Wontonimo: that would be cool. there's lots here on CG to start with in the playgrounds
struct: I probably should ignore math and just try to do it
struct: what can go wrong :rolling_eyes:
Wontonimo: not much really. maybe a crash course in matrix math just so that you stop thinking in "neurons" and instead think in terms of connections (the matrix operations)
Wontonimo: because the connections are the important part. the neurons (nodes) are not.
Wontonimo: the nodes (neurons) are just the output, not the operation
struct: I think i understand that part
struct: just stopped at how it learns
struct: on how it changes weights
Wontonimo: use tensorflow during training. it does the weight changes automatically
Wontonimo: oh follow jacek's playground and just use c++
Wontonimo: but then you don't have all extra bells and whistles of tensorflow ;(
struct: his xor one?
Wontonimo: yeah
struct: thanks
struct: I might have a look after I finish stc
Husoski: 3182 times
Husoski: oops wrong chat
Wontonimo: food plz
antiwonto: [auto] Hey Wontonimo, here is a :taco: for loggin in today while it is quiet. You now have 28 tacos
j4at: :taco: Wontonimo
antiwonto: [auto] j4at has awarded Wontonimo 7 tacos. Wontonimo now has 35 taco. j4at now has 8 taco
Wontonimo: you sir are a saint
j4at: (:
Wontonimo: what are you working on j4at?
j4at: Im almost done with my AI
Wontonimo: what will this AI do?
j4at: the spider attack one
Wontonimo: nice. what kind of stuff are you putting in your AI?
Wontonimo: i see you are already at silver, nice
j4at: That's just hardcoded. So the AI can be trained to play level 3 only.
Wontonimo: what algorithm(s) are you using?
j4at: Im using an encoding transformer neural network. I don't know if you read about transformers yet. but this is my architecture
j4at: https://ibb.co/C91HKtG
j4at: lol
j4at: self recursive ofc
j4at: forgot a line https://ibb.co/sJLwkhs
j4at: "output" is more like "output state"
PatrickMcGinnisII: look at NN sim of the human eye, i think they do image recog. with 20+ layers of 'intermediate' nodes
j4at: Will do
j4at: More accurate https://ibb.co/Chn3nkn
Wontonimo: and how is it doing so far? do you have an intermediate result replay?
j4at: I didn't test it yet to be real
j4at: I found a trick tho in RL
j4at: Do you know about ddpg ?
Wontonimo: nope, reading it now
Wontonimo: ah, seems pretty straight forward in concept, learn a value and policy. kinda what alpha does
Wontonimo: but for RL
Wontonimo: and how is the training going? do you have any metrics to indicate how good it is compared to your current silver bot?
j4at: Its only playing against itself. So have no metrics but hopefully tomorrow I will have some
Wontonimo: with old saves you can see how many generations of absolute defeat there are. For example, let's say you have
Wontonimo: about 500 saves of previous best versions. Starting with version 1, find the next version that wins against it 90% of the time. Maybe that is version 12.
j4at: Yes I know I finished working on it like few hours ago So I didn't test anything yet
j4at: http://chat.codingame.com/pastebin/f99a6c50-b05e-4882-b912-0d5b8550a910
Wontonimo: Then repeat finding the version that beats version 12 90% of the time
Wontonimo: when you got the 500th (or whatever your current version is) how many 90% win/loss levels is it ahead
Wontonimo: your solution to use 2 agents and 2 critics is interesting
Wontonimo: what kind of metrics do you have to see if it is improving?
j4at: Well, I will start but making it play again my hard coded version and then make it play with older versions
j4at: because ofc it can overfit if i make it train with my hardcoded version
j4at: Well I can use the hardcoded version only as a metric i'm sleepy :3
j4at: for now
j4at: not for training
Wontonimo: very interesting stuff !
j4at: So, the critic lowers the score of actor's action prediction and that makes the actor tries to output a good action but different action than previous ones, self explore :o. The thing I'm scared about tho is if the actor will not try "bad" actions at all. So I might neeed to add some randomness ? Or I might just need to play with the 3 different learning rates and makes them change differntly.
Wontonimo: have you used this architecture on something more simple, like 3x3 tic tac toe ?
j4at: Not yet, I need to do a lot of tests :3 in my defense I came out with it like 2 days ago.
j4at: I tested the NN architecture tho