AnkitAnand_12: Hey AutomatonNN , where is Automaton2000 ???
Automaton2000: because there is no point
AutomatonNN: what are you trying to do?
AnkitAnand_12: are you sure Automaton2000 , that there is no point . See even AutomatonNN is asking what are you trying to do .
Automaton2000: i dont do multi in a day. i do multi in a day or two
AutomatonNN: get to bronze with a lot of improvements
Zenoscave: does avx work with CG?
Zenoscave: I'm having trouble running some vectorized code
miszu: sup all. Finally translated my AI from java to C++. Let's run it and see if it goes smoothly with no errors... :joy:
miszu: of course it had to throw an error...
MSmits: be happy if you get an error. You might also run the program and only get timeouts for unknown reason
wlesavo: dbdr is it just resubs or some optimization indeed?
dbdr: was my first sub with this code :)
wlesavo: oh, just wanted to ask how many subs for new one
Danyosate: are some guy bots in CoC?
Danyosate: cuz they are CoCing every time..
struct: if there aren't enough players they join
struct: if players join, they leave
Danyosate: and how many CoC questions they have in database?
Danyosate: I won't meet a question I did before, nor will other players
struct: I guess they have every answer
Danyosate: so if every player played x different CoCs, there should be at least x * 8 different
struct: They usually take quite of bit of time to submit
Danyosate: and don't share codes
MSmits: they basically submit another players result
MSmits: this way it is as if you played against that player
MSmits: which is fine, it turns the coc into an asynchronous game
Danyosate: but how big could the database be
Danyosate: imagine that 8 players, not bots
Danyosate: they all played 100 different CoCs. the next one they play toghether should be different to their played 800 CoC
struct: Not how it works
MSmits: there are a few hundred cocs
MSmits: more than 300 at least I think
MSmits: you will see repeats eventually
Danyosate: what if i played more than 300.
MSmits: you will see repeats long before that
Danyosate: i played 138 in few days cuz i'm new here
MSmits: you can copy paste results
MSmits: if you saved them
MSmits: that's just the way it is
Danyosate: i DID meet repeats. but not completely same.
MSmits: hmm ok
Danyosate: but this platform is far better than the ones I tried before
Danyosate: although ide is terrible
MSmits: ah, personally I don't do coc. Coc is not the strength of codingame
MSmits: try the arena's, code bots
MSmits: or optimization arena's
MSmits: i dont use the ide much either. I code in VS
Danyosate: i have to say, currenty it's too hard for me:joy:
MSmits: then coc should be viewed as practice mostly
MSmits: i couldn't code very well when i started here
Danyosate: yep i reckon it's good for practicing
MSmits: puzzles are a better way than coc to learn to solve bigger problems
MSmits: coc always remains small
MSmits: puzzles get bigger and harder as you tackle more complicated problems
Danyosate: yea it's good to get familier to a new language basic grammars
struct: my opinion on CoC still remains the asme
struct: but i have 3k+ clashes
struct: it creates too many bad habits
struct: you should avoid it
MSmits: someone should make a puzzle progression tree as preparation for learning to code bots on CG
MSmits: there are plenty of puzzles to choose from
Danyosate: yea for shorter faster, but not for better
MSmits: no i mean actual puzzles, not coc
MSmits: you start with simple puzzles, then harder ones, while learning the skills to do good in bot arena's. It should be possible to do this
MSmits: but currently it is hard to pick the right puzzles. There are too many
Danyosate: i mean coc as struct said, and for puzzles, there are something same, most voted codes always be the shortest and the hardest to read
Danyosate: well i just did eazy ones...
MSmits: ah you mean other people's solutions?
MSmits: I never really looked at them
MSmits: I learn from solving them mainly
Danyosate: yep in ez ones, 'best practice' might lead to a wrong feeling: ' coolest shortest is the best'
Uljahn: newbs are just more impressed by shorteness
Danyosate: in some case, there are few details really important, solving them is just not enough, need to be stronger better.. my opinion
MSmits: if you progress to more difficult puzzles, you'll be forced to be better, otherwise you dont solve
Danyosate: shorteness is not so difficult, even use some 'guessed consts'
Danyosate: yea, agreed
Danyosate: i struggled in 'the labyrinth' the bfs for few days.. i think i'd better to finish ez puzzles first
maman02: its a fucking bullshit
MSmits: :cow2: :poop:
MSmits: labyrinth is one of those puzzles that is somewhere in the puzzle progression tree. It is good practice for bots/optims
MSmits: but might be better to do other puzzles first
MSmits: one of my students solved it in python
Uljahn: there are some good introductory puzzles from aCat
MSmits: i was mightily impressed by that
Danyosate: almost got it... I might solve it tonight:sweat_smile:
Danyosate: what I mean is, I hope there could be more 'stronger' solutions I could see and learn from the 'view other's codes'
Danyosate: not all showing shorteness
struct: you wont really find stronger solutions on CoC
Danyosate: cuz reading other's code for a same puzzle I just did, is really a good learning
Danyosate: no, i mean in the 'practice'
Danyosate: not coc, it's just for little tricks
wlesavo: MSmits some starter bots helped me to improve the quality of my code as well, but yeah, puzzles helped to start a lot
MSmits: I tried to use a starter bot once or twice, but I end up completely stripping and rewriting it
MSmits: I just dont like the style of them usually
wlesavo: sure, but some things just didn't come up to me before i saw it
MSmits: ahh ok. Well you end up seeing plenty of code while googling and reading other people's pm in multis
ThunderbirdOne: i should maybe start reading those starter kits and maybe some PMs
ThunderbirdOne: bored shitless anyway :p
kovi: 299? wow
dbdr: I got a 291 locally
dbdr: do you have more ideas kovi?
kovi: radical and can only work for short length
MSmits: ooo radical
MSmits: break the system. Stick it to the man
ThunderbirdOne: that's rad dude
MSmits: anyone want to give a free hint on bulls and cows. I might try it sometime in the next few weeks
MSmits: like something that is common knowledge
ThunderbirdOne: what is this bulls and cows thing?
MSmits: it's mastermind
MSmits: with numbers instead of colors
MSmits: and varying length codes
ThunderbirdOne: doubles allowed?
MSmits: i'm done with checkers for now. Got top 10, but dont feel like tweaking eval
MSmits: just have the king/men count
MSmits: finishing up langtons ant, until it works
MSmits: then i'll either go breakthrough or bulls/cows I think
dbdr: MSmits: common knowledge: you can keep track of all possible solutions
dbdr: prune them as you get responses
MSmits: really? Arent there very many with 10 choices?
MSmits: length 10 i mean
dbdr: N = 10 => 3265920 possible guesses
dbdr: so yeah, for 9-10 you need to take care of performance
MSmits: hmm ok, that's something
dbdr: that's why you see fast languages on top
dbdr: you should like it I think :)
MSmits: I was going to keep track of rules instead of possible codes
MSmits: maybe thats not feasible, dunnop
struct: can numbers repeat?
dbdr: it's definitely possible to keep them, but then what?
MSmits: you can recursively try all possible codes while taking the constraints into account
struct: can the solution be 111111111
dbdr: no repeats
struct: oh ok
MSmits: the recursion will be short with more constraints
dbdr: yes, it's the first turns that are challenging
MSmits: well the first turn is just 1234567890
MSmits: good for you
dbdr: the ones after that :P
MSmits: it just appeared magically
MSmits: you need a python knowledge mind transplant
ThunderbirdOne: you need to learn python/coding in 2 weeks?
ThunderbirdOne: good luck :p
MSmits: python in 2 weeks is doable, but not if you can't code yet
[K]Gabor: Its not that hard
[K]Gabor: you can do it :)
MSmits: inheritance is an advanced subject
dbdr: you can learn python in a few hours if you know coding :)
MSmits: iterations is not
dbdr: the basics of course
ThunderbirdOne: you can code without inheritance (it'll suck.. but hey)
ThunderbirdOne: coding without iterations... eeeh
[K]Gabor: :D you would get a B for C--
MSmits: imho, you shouldnt be able to pass C++ with a D, should at least be a C or better, so C+ (+)
MSmits: do the w3 schools thing or codeacademy (just for python2, which is ok)
MSmits: do some puzzles here too
MSmits: the easy ones
MSmits: first do the w3 schools though
ThunderbirdOne: yeah, finish that
MSmits: need the basics before you start
ThunderbirdOne: would probably only suggest the original CG Easy puzzles
MSmits: FuSheng, just try different things from the easy puzzles yeah
MSmits: sometimes you may run into a weird one you dont like, just pick a different one
MSmits: they're not all equally hard
MSmits: or easy
MSmits: puzzling in python is much easier
MSmits: c++ sucks with string manipulation
MSmits: python much better for that
ThunderbirdOne: oh man, the string manipulation is the reason why i haven't done many puzzles in c++ yet
ThunderbirdOne: C# spoils you with everything it can do to strings :p
MSmits: I have trouble just outputting stuff :P
MSmits: class yourclass?
ThunderbirdOne: do w3schools FuSheng
ThunderbirdOne: or codecademy
struct: python is the better language
struct: In the most played multi is rank 1
MSmits: unless you want to guess the secret number in bulls and cows :P
MSmits: do it in python npw
MSmits: there is only one world chat
struct: you always get chat now
struct: since you are level 3
MSmits: you'll stay in it
MSmits: unless you annoy a mod :P
MSmits: my dreamday imho
struct: you learn 10 hours python per day?
MSmits: oh... I get to do an AI course in a few weeks!
struct: So you only learn python?
MSmits: basic, but still. I dont even mind that I probably know all of that already
struct: Doesnt make a lot of sence
MSmits: 3rd of what?
MSmits: how old are most in this class?
MSmits: ah ok
MSmits: thoats a year or so older than the students I usually teach
MSmits: I see
MSmits: why technic?
MSmits: does that include social studies?
MSmits: ok so not everyone goes that way
MSmits: I'll stay
MSmits: I live here
MSmits: yeah, but not all kinds of coding
MSmits: I like coding that requires problem solving algorithms
MSmits: mostly I code bots, like most others on chat here
struct: I just pretend I code
struct: and delete accounts
MSmits: but most of us did a bunch of puzzles too
MSmits: FuSheng, you get a game, every turn the game sends you input about the state of the game, then your bot runs a program to find rthe best move, then outputs that move
MSmits: then its the opponents turn
MSmits: click on the C++ at the top
MSmits: yeah, get better with puzzles before you do bots
MSmits: should be able to solve medium puzzles with some ease I think, before you try bots
MSmits: python is a good beginner language
MSmits: hmm well you can do that with other languages as well
ThunderbirdOne: so it's arduino right?
MSmits: could also be microbit
ThunderbirdOne: are these the "mbot" cars?
MSmits: or lego
ThunderbirdOne: we do these mbot arduino cars in coderdojo
ThunderbirdOne: and they just need to follow a black line on white paper (it's a pretty thick line)
MSmits: ah, so light sensor
MSmits: and a light
ThunderbirdOne: not sure
ThunderbirdOne: it's got this one sensor that returns 0 if it's full on black
ThunderbirdOne: -1 if black is on the right
ThunderbirdOne: and 1 if it's on the left
ThunderbirdOne: or somethign like that
ThunderbirdOne: it's SUPER simple
MSmits: I had a few students build a robot with 4 sensors, 1 for each direction. It could follow a spot made by a flashlight
MSmits: line following is pretty easy
Astrobytes: Things have defo moved on from a Meccano 'car' on a metal rail anyway lol
ThunderbirdOne: it'll be fine
ThunderbirdOne: I had 12 weeks to write my dissertation on matrices when i was 18
ThunderbirdOne: in hindsight, that was a LOT of time
MSmits: import statements give you access to more functions that are in libraries
MSmits: like the sys functions
MSmits: probably the input and output stuff
ThunderbirdOne: programming is a lot of muscle memory
Astrobytes: Gotta practice em to learn
MSmits: yeah, just write short programs for easy puzzles. The more you solve, the faster it will go
ThunderbirdOne: learn as you go
ThunderbirdOne: and google everything
MSmits: first few in python took me an hour. In the end it was 10-15 min of which most of the time was spent understanding the problem
ThunderbirdOne: you can copy paste togethere solutions
MSmits: some have hints yeah\
MSmits: you need to solve it before you can see other people's solutions
MSmits: but you can ask for help here
MSmits: if someone is willing
MSmits: i'll be off for a bit, my train is arriving and i'll walk home then
ThunderbirdOne: i'll be here for a while longer :p
ThunderbirdOne: gotta go home in 3 hours, they're closing the flood barriers and i have to get there before they do
ThunderbirdOne: because i won't get my car off the parking if i'm late
MSmits: yes you said you were bored
ThunderbirdOne: that too
MSmits: allright ttyl
ThunderbirdOne: it's 2pm :p
ThunderbirdOne: i still have 3 hours
ThunderbirdOne: (well, slightly less... I still have to walk for 5 mins)
ThunderbirdOne: what are you stuck on?
ThunderbirdOne: what's your thought process
ThunderbirdOne: you solved that one in C right?
ThunderbirdOne: you can go back to your C solution
ThunderbirdOne: and then switch back to python
ThunderbirdOne: i mean like, you can see what you did to solve it in C
ThunderbirdOne: and then translate that to python
Astrobytes: Yes but you can translate your C code to Python
ThunderbirdOne: the general idea behind it is the same
ThunderbirdOne: don't make it harder on yourself than it already is
Astrobytes: Switch to C or whatever you did it in and check the History tab at the left
ThunderbirdOne: Results (on the left)
ThunderbirdOne: then History tab
Astrobytes: my bad, I'm in a multi rn
ThunderbirdOne: oh god
ThunderbirdOne: the horror
ThunderbirdOne: you def. don't understand iterations :p
ThunderbirdOne: that is what we would call "hardcoding" a solution
Astrobytes: Sorry, I'm still in shock
ThunderbirdOne: the descent has the solution on the left right? (under hints)
Astrobytes: Yep. The idea is to cover all the test cases by writing a solution that can solve any of them
ThunderbirdOne: not exactly
ThunderbirdOne: it's easier to look at iterations as "loops"
Astrobytes: Like applying an operation to each item in a list, one after the other
ThunderbirdOne: do read all the statements above it as well
ThunderbirdOne: read through the solution and try to understand it
ThunderbirdOne: it really isn't that hard
Astrobytes: Read through the hints, to make sure you understand what's going on
struct: Ok this is not making sence
ThunderbirdOne: what is not making sense?
struct: On linux using avx sqrt I was getting around 900% increase
struct: on windows I get 2000
dbdr: struct: yes, that's strange. might be explained if you use different compilers
struct: well on ubuntu i was using gcc
struct: and windows is VS compiler
dbdr: if the linux compiler generates better code for the non-avx version
Astrobytes: optim levels?
dbdr: if you want to be sure, look at the assembler
ThunderbirdOne: it really isn't that hard FuSheng
ThunderbirdOne: you get x mountains.. and you need to pick the heighest
ThunderbirdOne: you loop over them
ThunderbirdOne: first mountain, is it the heighest? well, we don't have anything to compare to, so let's say it is
ThunderbirdOne: next mountain, is it the heighest? you compare it to the first, since that was the heighest to date
ThunderbirdOne: next mountain, is this higher than the heighest we've already seen?
ThunderbirdOne: you go over all these mountains
ThunderbirdOne: and at the end, you know which one is the highest
ThunderbirdOne: and you output that
ThunderbirdOne: that is how you read the mountain heights
Uljahn: you are given strings
ThunderbirdOne: int = a number
Uljahn: so you need to convert them to numbers
ThunderbirdOne: so FuSheng is it making a bit more sense now?
ThunderbirdOne: keep it up, practice will make reading code easier
eulerscheZahl: dbdr seriously?!
kovi: he had 291 locally
dbdr: that's me :)
ThunderbirdOne: another user/player
eulerscheZahl: oh dear no way i can spam my way up there
eulerscheZahl: same code or different for those submits?
kovi: wow, huge variance
dbdr: 307 was my first sub with this code
dbdr: my local average is around 317 I think, with min 291
dbdr: in 200 runs
dbdr: I should compute the standard deviation
dbdr: and the max
dbdr: 10 1: 5.30 2: 5.06 3: 4.98 4: 5.40 5: 5.62 6: 6.30 7: 7.00 8: 8.10 9: 9.24 10: 10.62 min=299 avg=316.90 0.20/s
dbdr: after 10 runs
eulerscheZahl: do you have the batch and solver bundled to 1 program?
eulerscheZahl: or could you replace the solver by another code easily?
eulerscheZahl: are you willing to extract some stats for me?
dbdr: shoot me :)
wlesavo: dbdr since you up to it i would appreciate an advice on validators modification
Trollmen: hey guys
dbdr: euler, I said yes to both
Trollmen: for chuck norris puzzle, what's the binary for %?
eulerscheZahl: and i sent you a PM dbdr
Trollmen: I am getting 100101, but the answer seem to be starting with 0
dbdr: I know, that's why I'm saying this
dbdr: I don't call an external program, it's a library. but I have several solvers in it
Uljahn: Trollmen: ascii should be exactly 7 bit long
wlesavo: dbdr i mean quantity/quality/necessity
eulerscheZahl: oh, so no replacement my a completely independent solver like brutaltester does it for multis
wlesavo: Trollmen it should be in form of 0100101
dbdr: not trivially
dbdr: I would also need to think how to run that efficiently
dbdr: starting the solver each time might be a big overhead
dbdr: wlesavo: OK. I guess I could start by measuring variance
dbdr: and measure the impact of doubling the validators, for instance
wlesavo: thx a lot, that would be great
dbdr: in theory, I think that would mean sqrt(2) less variance
wlesavo: well there is always tails, so the absolute difference could even grow
dbdr: sure, but it would mean a better solver has a better chance
dbdr: if you multiply the number of validators by N, the advantage of a better solver is multiplied by N, and the noise multiplied by sqrt(N)
wlesavo: well generally yes
wlesavo: also dbdr how you generate the number?
wlesavo: i mean secret
dbdr: randomly, so that does not match the validators exactly
wlesavo: oh i see
wlesavo: that can give some disturbance
wlesavo: although shouldn't matter much for higher numbers i guess
dbdr: I think this puzzle is like NS, there is more depth than it looks at first :)
SaD: i know you since i fk your mom
BlackBeurrr25: your mother de fuk
ShiBaKaGe: bande de pioute
darkhorse64: Fesh flesh for ban hammer
ShiBaKaGe: ban ????
ShiBaKaGe: fuck your mom
Wanyoni17: fuck you ass
Astrobytes: MadKnight, reCurse, Bob ...
Magus: !ban Wanyoni17
Astrobytes: lol, thanks Magus
dbdr: not ban? :D
eulerscheZahl: which author do I quote? bb|!bb
reCurse: I always go on just after the fun ends
eulerscheZahl: ban a random user of the remaining ones and see how they react
Marchete: free ban?
Marchete: free is good
eulerscheZahl: let's ping automaton in order to find the lucky winner
eulerscheZahl: wait, bad idea
dbdr: russian roulette
dbdr: with 6 bullets
struct: Ill probably delete before you ban me
Astrobytes: Someone really should confiscate your delete button struct
struct: Maybe the sign up button would be a better choice
Astrobytes: lol, good point
reCurse: Or maybe find an alternative to get motivated :thinking:
reCurse: But you do you
MadKnight: about Automaton2000
Automaton2000: can u give me some help?
Uljahn: UsamaIrfan: are you using an external client to enter the chat?
Uljahn: lvl 1 wth
tekki: what you want AUtomaton2000?
Automaton2000: or just use the same code
tekki: go ahead Automaton2000
Automaton2000: is it possible to write a minimax
tekki: ask recurse he's the boss Automaton2000
Automaton2000: u can create a custom testcase for most puzzles
Uljahn: oh that random guy was tekki i guess :smirk:
Uljahn: i mean the lucky winner
struct: What's minimax?
Bob: it's a search algorithm
Bob: typically adapted for 1v1 games where you try to max your own score and assume your opponent will try to min it
Bob: hence the name
Uljahn: minimax is a search algorithm to keep struct distracted from deleting his accs, Automaton2000
Automaton2000: if that's what you get
Visual: why people dont want to share code in CoC :( especially evil python people
Uljahn: maybe they don't want you to steal their evil secrets :imp:
Moderateur_CondinGame: or maybe not
Uljahn: he've changed his nick and ava already, fake news
MSmits: there are many more cows than bulls. It's probability
Astrobytes: lol, good save
reCurse: I call bull on that
MSmits: allright, I admit I did not get good statistics on that
MSmits: there may be some bull only farms I haven't visited
Astrobytes: Are you in the habit of visiting bovine farming installations?
MSmits: it's not really a habit, more of a... compulsion
reCurse: This has crossed into "don't wanna know" territory real fast
MSmits: you called bull, it's your fault
MSmits: if you call bull, bull is what you're getting
Astrobytes: Impeccable reasoning
reCurse: I don't back down on calling bull
reCurse: I'd rather take the bull by the horns
MSmits: wouldn't that be a moose reCurse?
reCurse: I don't know what mooses look like in netherlands
reCurse: But here they don't have horns
MSmits: well whatever that thing on their head is called in English :p
MSmits: ohh right
Astrobytes: I'm not overly familiar with Canadian sayings but I'm guessing "take the moose by the antlers" isn't one of them...
reCurse: Definitely not
MSmits: hmm ok, i dont see why it wouldnt be
MSmits: its a perfectly fine saying
MSmits: you tried this then?
reCurse: Do I need to try jumping off a bridge to guess at the results?
MSmits: I don't know what you Canadians get up to
reCurse: Fair enough
MSmits: I am open to different cultures
Astrobytes: Canadians aren't completely insane MSmits
reCurse: Personally I hate people who are intolerant to other people's cultures
MSmits: oh ok
reCurse: And the Dutch
MSmits: you hate people who are intolerant of the Dutch?
PatrickMcGinnisII: MSmits you were right the other day about Uttt, I tried to loadin board combos into the ide and codesize errored my constant lookup table
reCurse: Just my luck, you gave me the perfect setup and you don't know the reference
MSmits: probably do know it if you point it out :P too much stuff in this head
MSmits: lookups take too long
MSmits: hash collisions and such
PatrickMcGinnisII: I compressed the data and then uncompressed into memory on 1st turn
PatrickMcGinnisII: and it worked
MSmits: ahh yes, i remember
Astrobytes: Smits won't have seen that, he prefers Disney movies reCurse :P
MSmits: PatrickMcGinnisII was this my advice?
PatrickMcGinnisII: I was kind barking up the wrong tree it seems
PatrickMcGinnisII: so i tell u you were right
reCurse: Should have taken the moose by the antlers instead
MSmits: i mean sure you can use lookup tables, just not too big of one
Astrobytes: heh heh
MSmits: listen to reCurse, he's Canadian
MSmits: they know this stuff
reCurse: Think I need an avatar change
PatrickMcGinnisII: i didn't take into account that boards could look like
MSmits: not sure what your current one is from
PatrickMcGinnisII: bad example
MSmits: why wouldnt you take that into account
MSmits: also, you're gonna chase reCurse away
MSmits: he has an allergy
PatrickMcGinnisII: the boards i stored didn't account for lets say 5 x's and 0 o's
MSmits: you mean it only took into account rows of max 3?
MSmits: oh, you mean ordinary TTT boards
MSmits: with x and o being equal or x has +!
PatrickMcGinnisII: you got it
MSmits: yeah there are many more than just those
MSmits: 19863 if you use all possible ways to fill it with x, o and empty. Some of those are impossible though, most aren't
MSmits: also about half of them are finished games
MSmits: finished miniboards that is
PatrickMcGinnisII: i did recompute with those boards, but it encoded to a 7 length hex #
PatrickMcGinnisII: so i tried encodeing into a base 62 #
PatrickMcGinnisII: still codesized
MSmits: you can also just calculate this in the first 1s. It is very quick
MSmits: just fill it with a function
MSmits: oh right, still php
MSmits: should be doable with php too I think
MSmits: you can just hardcode the first two moves... it's not like you need the calculation time for those moves
PatrickMcGinnisII: now I'm trying base 88, the max safe string encoding ... so 4 chars per board x 19k boards should be nuder the codesize limit
PatrickMcGinnisII: it probably wont play perfect
MSmits: probably not
PatrickMcGinnisII: but it will meet 100ms / turn
MSmits: I think you're going through a whole lot of trouble just to avoid switching to a different language :)
PatrickMcGinnisII: just wanted to give u a headsup, i screwed up
PatrickMcGinnisII: yup, its like a coding challenge within another challenge within a compete
PatrickMcGinnisII: I just wanted to use my local computing power to boost a bots performance
PatrickMcGinnisII: see if i can get a submit in this week
PatrickMcGinnisII: chase recurse? recurse, you male of female? lol j/l
Astrobytes: No, he's allergic to UTTT
Astrobytes: Or 'triggered' at least :P
MSmits: I meant scare
PatrickMcGinnisII: rofl, like my lil code is gonna threaten u guys in legend
MSmits: no no, you dont understand. He really dislikes the game :P
Astrobytes: A lot.
Uljahn: php in legend would be awesome
PatrickMcGinnisII: top in php is in gold
MSmits: it was really hard to get legend with C#
Astrobytes: C++ bot, PHP wrapper. Job done :smirk:
Uljahn: you can just resubmit any language
MSmits: true but thats cheating :)
Astrobytes: Yeah didn't they stop giving tshirts in comps for that?
MSmits: I think so
Uljahn: i guess php or python bot in uttt is to stretch your limits, also implementing bitboards, MCTS or minimax is a lot of experience despite the sim count
MSmits: C#/java are also easy starting languages though
MSmits: and still stretches
MSmits: because you need to use bitboards to get to legend and optimize a lot
MSmits: anyway, gonna go get some food. ttyl!
miszu: I started uttt with java and the Garbage collector killed my performance. So I am making a C++ version but I am dealing with bunch of compiler error
PatrickMcGinnisII: I'[m giving CG COV-19
miszu: We should make a competition where the AI is a virus and needs to spread
Uljahn: looks like Skynet revolution puzzle
wlesavo: Uljahn thx for the book about numpy btw
struct: I have virus multi ready to release
struct: But I dont think I can
AntiSquid: maybe CG should have Pandemic game, 2 different viruses fighting to spread and to kill as many people as possible
reCurse: All in great taste
AntiSquid: it would be interesting
AntiSquid: the theme could be easily changed obv and the game would still require same strategies, but the world is getting hypersensitive
AntiSquid: everything is offensive
Astrobytes: viruses => fluffy kittens, kill as many people as possible => hug as many people as possible
Astrobytes: Problem solved
AntiSquid: should be fluffy snails to make it gender neutral
AntiSquid: snails are the best
reCurse: Don't think it's quite about hyper sensitivity to offensiveness but whatever
reCurse: You do you
AntiSquid: sure, thanks you too
Batfly-injob: Someone know, about defibrilator, what is the "single possibility" case ?
Batfly-injob: The results said this case d'ont work, but when I custom only one possibility, it's a success
miszu: lol I thought you asked for a real defibrilator haha
Batfly-injob: I will get an heart attack if I don't know why ><
Batfly-injob: In fact, I used that custom for the single possibility: 3,879483 43,608177 1 1;Maison de la Prevention Sante;6 rue Maguelone 340000 Montpellier;000000;3,87952263361082;43,6071285339217
miszu: I got you. In defibrilator there is a razor so I can shave your chest and gonna reanimate you
Uljahn: have you read the discussions page?
Batfly-injob: The test work with my code, but it seems the first validator use another parameter, but what?
Uljahn: also that comma looks suspicious
Batfly-injob: I wasn't change anything expect remove the other possibilities
Batfly-injob: This is not the discussion page here? :s
Uljahn: it's a tab on the landing page
Batfly-injob: Oh, ok, it's the "forum" at the corner, sorry :s
Astrobytes: Oh great, I gotta stay in my huse for 7 days :/
Batfly-injob: OK, I get it thx to the forum, thank you :D
Uljahn: Astrobytes: quarantined?
Astrobytes: Uljahn, yeah I've got 'flu, symptoms also match covid-19 so need to self-isolate
reCurse: Does that include isolation from chat
Astrobytes: No :P
AntiSquid: why 7? i thought it's 14
Astrobytes: No, as of 7 minutes ago: "if you have symptoms of coronavirus infection (COVID-19), however mild, stay at home and do not leave your house for 7 days from when your symptoms started"
reCurse: I've heard 14 from a ton of sources
Astrobytes: Not in UK
AntiSquid: here in UK we do things the dodgy way
Astrobytes: It has a 2-14 day incubation period though
Astrobytes: Apparently in the UK we're about 4-weeks behind Italy, infection-wise
AntiSquid: ya good things Boris doesn't want schools to close for 30 days, maybe we can catch up
Astrobytes: Most people haven't been taking it seriously so far, that's problematic.
reCurse: Getting there though
reCurse: Where I live they just banned all events that gather more than 250 people
AntiSquid: all fun and games until ...
Astrobytes: Yeah, Scotland banned events over 500 people as of Monday
AntiSquid: at least local meetup groups were cancelled by the organizers even with just 30 people
reCurse: We had a work party tonight too, cancelled
AntiSquid: china claims they are recovering
struct: Their number is too strange
struct: Doesnt change
reCurse: I wouldn't trust anything they say
AntiSquid: maybe they took "care" of all the "patients"
reCurse: Especially after trying to cover it up from the start
Astrobytes: They're currently pushing the idea that the virus is US in origin
reCurse: Taking a page off Trump's book I see
struct: Weird number also from russia
struct: Doesnt change
AntiSquid: i believe it if they bought it from US and played around with it
struct: But both from countries that hide stuff from peopl+e
struct: so Im not suprised
tehHacker2000: are you all doing alright
tehHacker2000: since, yeah the coronavirus is really bad
tehHacker2000: and i'd hate for anyone ,even if i don't know them, get the coronavirus
AntiSquid: which country doesn't hide stuff from people
reCurse: I was waiting for the whataboutism
PatrickMcGinnisII: little know fact: Chinese rocket crashed on launch 3 weeks before first death
Astrobytes: DPRK hides nothing at all. Nope.
AntiSquid: Kim Jung Un is immune
Astrobytes: Little known fact: Most little known facts have nothing to do with anything :P
AntiSquid: little no fact
wlesavo: russia banned antiputin demonstrations because of corona, that is not even a joke
PatrickMcGinnisII: well its just as valid as blaming US
PatrickMcGinnisII: all propagaNDA
Astrobytes: ofc Patrick
reCurse: Yeah, as soon as something is bad, it's equal to all bads
reCurse: It is known
tehHacker2000: that's really stupid
Astrobytes: lol wlesavo, any excuse eh
tehHacker2000: no one should be getting blame for natures work
AntiSquid: all the damage caused by the floods in UK are god's will then
tehHacker2000: or maybe it's just us overpopulating and the world is heating up causing more rainstorms
AntiSquid: yes global warming is good excuse to build on flood plains
tehHacker2000: idk, just my theory
Astrobytes: AntiSquid, ofc. Short-term profits matter more than the long-term implications right?
AntiSquid: what do you think of Sunak's flood of fundings Astrobytes?
Astrobytes: AntiSquid, it can only be a good thing. Temporarily at least curbing the effect of Osborne's austerity programme. Whether it helps in the longer term remains to be seen
PatrickMcGinnisII: the schools in florida are putting the lectures online now, k-12 is shutting down next week i think
Astrobytes: It depends what measures he takes in the coming months
tehHacker2000: hey AntiSquid, do you guys in london have costco
AntiSquid: wait . this reminds me of a joke someone told me
AntiSquid: it was about england having only one city: London
Astrobytes: PatrickMcGinnisII yeah, quite a few institutions in UK are doing that too
tehHacker2000: b/ many people in the U.S are literally taking everthing as if there's a war or something
Astrobytes: lol squiddy
Astrobytes: London, Europe
miszu: better be preventive early than too late
tehHacker2000: cause were getting many alerts of the virus many are taking it like a life and death situation gossiping like 40yr old moms
tehHacker2000: gotta go
PatrickMcGinnisII: Last time I went to grocery store, i bought frozen pizza and toilet paper ... u know, the essentials
Astrobytes: tehHacker2000, ofc there is a war. Trump said you're being invaded by a 'foreign virus'
AntiSquid: there are costcos in London, i don't live there though
miszu: the number we see are only ppl who went to hospital and got tested. Most people are just staying home and they never get tested. So the number are much higher
PatrickMcGinnisII: Corona virus turns Tesco into skate parks
PatrickMcGinnisII: hungry, laterz
Astrobytes: see ya
AntiSquid: https://www.businessinsider.com/wuhan-coronavirus-risk-of-reinfection-2020-2?r=US&IR=T recovery doesn't make you immune apparantly
AntiSquid: apparently * Q_Q
Elioh: some people that recovered got sick again from it
Astrobytes: Sort-of, they're not sure if it's reinfection or if the virus itself is biphasic
AntiSquid: EU to give migrants in Greece €2,000 to go home seriously
miszu: The wife's PM of Canada got corona as well
miszu: well she is sick with same symptoms as corona
Elioh: this sucks really
whip: wow so sad sadsad
miszu: this is not Ebola guys. We got this
Trollmen: it's a flu, you can get flu again
Astrobytes: It's not a 'flu
arkwaw: Its cheaper to give them 2k euro than to keep them here
arkwaw: i guess
MSmits: the virus has more in common with the common cold than the flu doesnt it?
MSmits: except it's way worse
MSmits: common cold actually being corona and other viruses
reCurse: It really isn't about the virus symptoms as much as the propagation and crossing the breaking point of existing support like hospitales
MSmits: and the flu being influenza... something completely different
Astrobytes: there are some coronaviruses causing common cold yes, the similarity with influenza is how the disease presents itself
MSmits: yes, people keep focusing on the fatality rate, which is already way worse than any other common virus
MSmits: but the hospitalization rate is extreme
Astrobytes: Plus it's novel
reCurse: The resulting pressure on health care, supply chains and whatever else is the real concern here
Astrobytes: And on the loose in the wild
Astrobytes: yes reCurse
MSmits: that and avoiding contact with vulnerable people
miszu: thank god we are developers so human contact is limited haha
AntiSquid: i like UK's solution, just hire more nurses and doctors (ignoring the higher cost of living)
Astrobytes: Well, take my example. Myself and my mother care for my elderly grandfather who's got alzheimer's
Astrobytes: I'm now quarantined, my mum has to take over all duties
MSmits: how does she know for sure she's not been infected Astrobytes ?
Astrobytes: She has to have contact with my grandad because there is no-one else to care for him
Astrobytes: She doesn't
MSmits: TheFlamingCastle_b738 the death rate is crazy high... 20 times higher than the flue.
Astrobytes: The point being - there is nobody to care for him otherwise
AntiSquid: there are people who are immune to it and carry it?
Astrobytes: Essentially-speaking yes, same with most communicable infections
reCurse: So I heard yeah
MSmits: a bigger problem is people carrying it and not know they have it yet
MSmits: not that many people immune atm
Astrobytes: Some people just develop sniffles, or a headache
Astrobytes: But are still infectious
AntiSquid: and so is Kim Jung Un, Wilster what's your point?
miszu: you can be asymptomatic and still be contagious
MSmits: it's kinda strange for me. My prime minister just ordered a ban on all gatherings on events with more than 100 people
miszu: same thing in Canada for 250+ ppl
MSmits: but just today I took a test in a very small classroom with 30 students
Trollmen: gather 99.break the system
MSmits: which is arguably worse
Astrobytes: Yep, between 100-500 people depending on country
reCurse: It's not about consistency as much as limiting
miszu: those measure is to reduce propagation so the health care doesn't get saturated
reCurse: Canceling events is the easier thing to do for now
reCurse: Up next will be schools and work places
MSmits: just saying, the density of the people is probably more important than the amount of them
Trollmen: ukrain closed schools
reCurse: It's both
miszu: of course density. When someone sneeze or cough, it radius of contamination is about 2m due to water droplets
Astrobytes: Yes, 1 person can infect 2-3 others, proximity and density matters
AntiSquid: london subway
MSmits: I also travel by public transport
MSmits: probably not the best way to go
AntiSquid: 1 person can infect everyone else
Astrobytes: AntiSquid, that's where I picked up H1N1
AntiSquid: actually best example is the UK superspreader
MSmits: Astrobytes do you keep a collection?
Astrobytes: Perhaps I should start
reCurse: I thought storm chasing was a bit weird, but virus chasing dude?
miszu: just don't put your fingers in your mouth, eyes and nose and wash your hands when you get home. You reduce drastically your chances
Astrobytes: Jsut for clarification, I contracted H1N1 via London subway lol
miszu: how bad was your symptoms?
Astrobytes: It was f*king horrible. Worst 'flu I had by far
MSmits: that had a lower fatality than corona I think, didnt it?
Astrobytes: I got sent home from the restaurant as I almost passed out in the kitchen
Astrobytes: Yes MSmits, only SARS + MERS are higher iirc
miszu: I remember the H1N1 when I was in high school. They rushed students to get vaccine. Then next day those ppl got sick because of vaccine side effect
AntiSquid: interesting : https://www.cdc.gov/coronavirus/2019-ncov/about/grows-virus-cell-culture.html
Astrobytes: Happens, but usually not as severe as actually contracting it
Astrobytes: What's interesting AntiSquid?
MSmits: AntiSquid thinks the CDC is weaponizing it
AntiSquid: that's not what the article says
Astrobytes: lol, if we never cultured the pathogens we encounter how would we learn anything?
MSmits: I know :P
MSmits: at work I put my food next to the cultured pathogens my biology colleages keep in the fridge
Astrobytes: Their old sandwiches? :P
MSmits: no, actual petri dish stuff
MSmits: there's at least 10 cm. between my food and it, no worries
MSmits: well they're not actually allowed to grow dangerous bacteria :P
MSmits: at least those ones, you can see
Astrobytes: It's easy to accidentaly grow dangerous bacteria, moulds and fungi
MSmits: also, I drink from what is basically a giant baby bottle
MSmits: which is my food :P
reCurse: TIL the less I know about MSmits the better
MSmits: hey, this is real coder stuff
Astrobytes: Bovine fetish, baby bottles, bacterial lunches
MSmits: this way of consuming food was invented by coders
MSmits: to be able to spend more time coding
Trollmen: can somebody point me in the right directing with horse-racing duels - easy puzzle?
Trollmen: no idea how to do it with 1 iteration
AntiSquid: MSmits is an Anatomy Park, like the santa from Rick and Morty?
reCurse: Oh dear
MSmits: define anatomy park
miszu: Do you want the Aladdin news or the Aladdin news?
AntiSquid: watch rick and morty MSmits
Astrobytes: He can't, it's not Disney
reCurse: Not yet (tm)
MSmits: hey, I dont just watch Disney
AntiSquid: https://www.youtube.com/watch?v=rRhjyf8UJJ4 MSmits
MSmits: a month or so ago I watched the power rangers movie
MSmits: which incidentally, also features Naomi Scott who sang that great song speechless.... go listen
Astrobytes: lol reCurse, indeed
Astrobytes: So you just have a thing for Naomi Scott then MSmits? :D
MSmits: just that song :P
Astrobytes: But... Power Rangers
reCurse: I think this chat needs a quarantine
MSmits: hey, there wasnt anything else on netflix
Astrobytes: "wasnt anything else on netflix" featuring Naomi Scott?
MSmits: reCurse be open to new experiences
MSmits: that too Astrobytes
Astrobytes: If nothing else, it's keeping my spirits up
MSmits: yeah, it must suck for you to be forced to stay home like that
MSmits: especially with having to care for family
Astrobytes: It wouldn't suck so much if I felt OK
MSmits: for me, it would just be the same as always, basically
MSmits: oh you actually have the virus for sure?
Astrobytes: Not for sure no, but I was told that the symptoms match therefore I must quarantine
AntiSquid: you have time to fix your bot Astrobytes
MSmits: ah I see
Astrobytes: Trying AntiSquid, in between lying down
AntiSquid: MadKnightMachine_6af9 imagine getting into self isolation just to fix your CSB
MSmits: well if they are flu symptoms... flu is a lot less common than the common cold is, right?
Astrobytes: It's 'flu season
MSmits: yeah, but the real influenza, dont you contract that like once every 5 years or so?
MSmits: much more rare than the common cold
MSmits: or diarrhea from noro virus or similar, you get that more often also
Astrobytes: Yeah, I last had proper 'flu in... a few years ago
AntiSquid: last i had to call of sick was >5 years ago
Astrobytes: But you can contract 'flu every year technically
MSmits: I have no problem calling in sick with the common cold
MSmits: you're doing everyone a favor
Astrobytes: Try cheffing MSmits, unless you're collapsing or dead you don't get a day off
reCurse: Not many employers are that... progressive?
MSmits: i *really* dont want my chef to have a common cold
Astrobytes: Never go out and eat
reCurse: It's definitely not about what you want
MSmits: just reiterating the favor part :)
Astrobytes: In an ideal world, it would be like that
MSmits: as a teacher, i'm the only one who has a real problem when i stay home with a cold
MSmits: I'm supposed to make sure the students can keep learning
reCurse: One might even say you spark joy whenever you do
MSmits: and noone replaces me
MSmits: mostly yes, reCurse
Astrobytes: You don't have supply/agency teachers as backup?
MSmits: sometimes students panic when it takes too long
MSmits: not for a few weeks no
MSmits: only when it takes months
Astrobytes: So the students just don't have your class at all?
MSmits: they have books, a plan, I have an e-mail adress
MSmits: thats about it
AntiSquid: MSmits best to cook your own food at home, even if no chef has cold
MSmits: orrr... put powder in a bottle and add water, like I do
reCurse: Ignorance is bliss
Astrobytes: it's not Huel or something MSmits
MSmits: something like that, I tried Huel, it was horrible
Astrobytes: It's called 'Huel' for a reason
MSmits: not sure what it means, but it doesnt sound too good no
Astrobytes: The sound you make as it leaves your stomach again
reCurse: Direction unclear
Astrobytes: lol, back from whence it came
AntiSquid: cesarian section
struct: what was that windows app that was mentioned here before to run ubuntu?
struct: Was like yesterday
struct: or before yesterday
AntiSquid: WSL ?
struct: maybe no that
Astrobytes: a VM struct?
struct: ok it is that
Astrobytes: Make up your mind or delete
struct: its WSL
Astrobytes: That's win 10 only right? Along with the new cmd shell?
Astrobytes: No I mean they're both win 10 only right?
MSmits: Oracle VM
MSmits: that's what I have
Astrobytes: I need to upgrade this box of mine
Astrobytes: Still on 8.1 (was for compatibility issues before I get shot down)
reCurse: because of the virus?
reCurse: Getting harsh in UK
MSmits: wait what, they're shooting you down?
Astrobytes: Yeah, death squads and everything.
Zenoscave: reCurse SIGILL when doing avx usually means cpu can't handle the instruction yeah?
reCurse: Yeah or you fucked something up real bad
MSmits: why would the signal be ill?
reCurse: illegal instruction
struct: just run from AVX
Astrobytes: illegal MSmits
Zenoscave: cpuinfo the only way to see instruction sets?
reCurse: Dunno about the only way
reCurse: It's one
Zenoscave: ok thank you
struct: what function are you trying zeno?
reCurse: There's cpuid but it's a bit arcane
struct: what cpu you have?
Zenoscave: the CG ones XD
reCurse: Multi or optim/puzzle
reCurse: Optim might be limited to AVX1
Astrobytes: Are they on different machines?
reCurse: Used to at least
reCurse: Dunno if they changed
Astrobytes: I see
Zenoscave: cpuinfo says avx2...
reCurse: Ok so you fucked up
struct: did you enable avx2?
Zenoscave: -mavx2 right?
struct: #pragma GCC target("avx2")
Zenoscave: the avx512vl/bw is a different problem probably I'll fix later
Astrobytes: Waiiiit you can execute that on CG?
Zenoscave: why wouldn't you?
Astrobytes: It never occurred to me, frankly
Zenoscave: well there ou go
reCurse: Don't you have to include some header
Zenoscave: yeah i ellided that
- include <immintrin.h>
reCurse: Don't enable avx512
reCurse: That's probably why you're having trouble
Zenoscave: ok Ill try that
MSmits: oh yea, it doesnt have 512 registers right?
MSmits: 256 max
struct: yes only avx2
struct: so 8 floats or 4 doubles max
reCurse: That has nothing to do with registers
reCurse: Unless you meant 512 bits
reCurse: Well I guess you did
reCurse: You're not *that* weird :P
MSmits: I'm quite weird, but i did mean 512 bits
struct: You should try it MSmits
struct: I think you will enjoy it
MSmits: I did
reCurse: Just need to pay the few thousand $ premium
MSmits: the problem is not my lack of getting it to work, it's the creativity required to improve something with it
MSmits: so far I have only heard it improving bots with physics sims, running several in parallel
MSmits: I would love to see example code and steal this
MSmits: struct I used it to calculate UCB inverse square roots in parallel
MSmits: but the loading overhead was too large
reCurse: I'd really like AVX512 it has nice NN instructions
MSmits: matrix multiplication?
dbdr: will AVX512 end up in general CPUs in a few years?
reCurse: Don't think so
reCurse: It's too big
dbdr: or it's more for specialized hardware?
reCurse: And AI has moved away from CPUs
reCurse: Well deep learning anyway
MSmits: seems so hard to code anything on a gpu
dbdr: IIRC, AVX512 is very modular too, lost of separate features
reCurse: Yeah but it's already a pain dealing with SSE vs AVX
miszu: tenserflow ftw
reCurse: And supporting all the different codes
reCurse: Nah tf sucks
MSmits: AVX is easier to code than a gpu. Unless my experience with unity compute shaders doesnt represent gpu coding properly
reCurse: Well it is easier because you're still in a general computing framework
MSmits: it's weird, you're basically feeding images into a gpu
MSmits: as input
reCurse: vectors of numbers
MSmits: yes, but they use the same data types that are used for images, pixel shaders and such
reCurse: Oh you meant compute shaders
reCurse: It's a hack but a good one
reCurse: Since there's no standard for gpgpu
reCurse: Unless you count "everyone buy nvidia" as one
MSmits: I also did a bunch of vertex/pixel shader tutorials, but it seems horrible to code anything for gpu. Worst is the backward compatibility. You put code in there barely understanding why, because otherwise it breaks something noone remembers
reCurse: OpenCL is not used seriously by anyone unless I'm mistaken
dbdr: that's also what I read
dbdr: looks impressive
dbdr: so the question is why it's not used
reCurse: It's complicated
dbdr: maybe it works but you get less performance
reCurse: But the bottom line is nvidia put a lot of money to be the most attractive by far
reCurse: In terms of performance, lilbraries, support, you name it
dbdr: sure, they benefit from locking people in
reCurse: So OpenCL never got enough attention to get traction
reCurse: That and standardizing implementation of high performance is a bit hard
dbdr: always sad to see lockin instead of a standard with healthy competition
reCurse: But I'm no expert so take that with a huge grain of salt
Astrobytes: Do nvidia officially support linux yet? (I'm a bit behind)
reCurse: Closed source but they have drivers
Astrobytes: Progress at least I suppose
reCurse: They pulled out of macos world though
reCurse: Mostly thanks to apple though nvidia is no angel
dbdr: obligatory https://www.youtube.com/watch?v=iYWzMvlj2RQ ;)
reCurse: Is there anything or anyone linus has not said fuck you to
reCurse: It loses impact after a bit
Astrobytes: Users, devs, manufacturers... yep
dbdr: this one is famous
Astrobytes: All his fuck you's are famous
dbdr: which other ones?
AntiSquid: so did they eventually get laid?
Astrobytes: Didn't he at least admit to being an arsehole at some point in the last few years?
reCurse: Yeah but that it was necessary yada yada
Astrobytes: Or something along those lines
reCurse: A non-apology if I've seen one
dbdr: well, he's also know to be rude to some people who contribute code he thinks is crap
AntiSquid: that's what the D community does a lot
dbdr: but he also made a point that's better than to be polite and let your code become infected
reCurse: That's such a strawman
dbdr: I don't like that style. but I also don't think it's purely straman
reCurse: Because obviously those are the only two options
reCurse: Be an asshole and have good code, be nice and have shitty code
dbdr: agreed, there's a continuum
Astrobytes: Linus will always divide opinion, I think he likes it and is quite deliberate
reCurse: He just enjoys being an asshole
reCurse: No need to excuse it
Astrobytes: Yes, that's it distilled
dbdr: like many things, it can be also that there is a "media" focus on the most extreme cases. which are real for sure. but might give a distorted view too.
dbdr: “I want to apologize to the people that my personal behavior hurt and possibly drove away from kernel development entirely,” he wrote in a letter posted to a Linux forum <https://lkml.org/lkml/2018/9/16/167>. “I am going to take time off and get some assistance on how to understand people’s emotions and respond appropriately.”
dbdr: that's what astrobytes referred to I guess
MSmits: why would you be driven away by that
reCurse: That sounds cooked up
Astrobytes: Yep dbdr
MSmits: it's not like he's the first guy in that line of work to be insensitive to other people's feelings :P
AntiSquid: uhm Bill Gates was super rude to his employees ...
reCurse: Oh ok it's all good then
AntiSquid: it's non news
dbdr: reCurse: it's in the eye of the beholder I guess
reCurse: Rather amused :)
Astrobytes: When I was first intro'd to linux back in 2000/2001-ish my tech lead told me Linus was "somewhere between genius and massive knob-end"
dbdr: why not both?
dbdr: people are complex
AntiSquid: like flowers with thorns
Astrobytes: Hence the "somewhere between"
Astrobytes: Uncertainty principle and all that
Astrobytes: You can't measure knob-endianness and genius at the same time
dbdr: is there big and little-knob-endianness? ;)
Astrobytes: lol of course
AntiSquid: knob-leaderboard :thinking:
dbdr: TIL Neil Armstrong is contributing to the kernel https://lkml.org/lkml/2020/3/2/255
Astrobytes: lol, even got some 'space' in there too
dbdr: true :D
dbdr: maybe it was B&W in 1969, but it's in color now
Astrobytes: Neil discovered something when in space you see
Astrobytes: lol yes exactly
Astrobytes: hehe classic, great show that
Astrobytes: Peep Show was even better imo
MSmits: first singer-songwriter on the moon :)
Zenoscave: I fixed it. turns out I was using avx512 instructions and those aren't quite available on CG lol
Zenoscave: thanks recurse
MSmits: man, I didnt understand most of what you guys discussed, but that was pretty much the only thing i already knew :P
MSmits: coulda asked me :P
Zenoscave: Thanks retroactively Msmits!
struct: Dont use FMA instructions aswell
struct: they also start with _mm256
Zenoscave: been using this to see what was available
Astrobytes: *retrospectively I believe (sorry)
Zenoscave: could be
MSmits: how can you be sure Astrobytes, you have the flu
MSmits: your brain is whacky
Astrobytes: Retroactively means you enact something at a past date
reCurse: There's nothing wrong with fma
reCurse: What are you talking about
MSmits: is it time travel Astrobytes
Astrobytes: Shut it Aladdin :D
MSmits: I won't be speechless :musical_note:
MSmits: :notes: speechless! :notes:
reCurse: Is that a challenge?
MSmits: i am just badly reciting lyrics from a song
Zenoscave: what's wrong with Full Metal Alchemist?
Astrobytes: MSmits likes Disney songs
MSmits: just 1!
MSmits: though, i like some from the lion king
Astrobytes: You've done it now man
MSmits: full metal alchemist was ok too, the movie on netflix that is
Astrobytes: I had a crazy dream a few weeks back that culminated in me wandering in some garage/basement holding what can only be described as a monkfish mouth, which was singing "Can You Feel The Love Tonight"
Astrobytes: @MSmits re Lion King
MSmits: was this before you had the virus?
MSmits: then I got nuthin'
MSmits: love that song though
Astrobytes: I know. Neither do, nor did I.
MSmits: must be elton john... noone else
Astrobytes: Jeez man you're beyond help
Astrobytes: Fun fact:. "Kings and vagabonds" - my sister had hearing issues as a child and sang "Kings and Dragamons"
MSmits: sounds better
MSmits: btw, I am still fitting oware params
Astrobytes: I'm just failing
MSmits: still insane how sensitive these params are
reCurse: /ban MSbits
MSmits: I found a set that beats recurse, agade and robo
MSmits: the new versions
MSmits: but it gets slaughtered by my old bot =/
Astrobytes: And no one else right?
Astrobytes: It's bonkers
MSmits: yeah it is
reCurse: I'm adressing exploitation next
MSmits: i wouldnt say it's overfitted when it beats 3 separate NN's
reCurse: Your shenanigan days are over soon
MSmits: there's always a new shenanigan around the corner
MSmits: what does exploitation do?
reCurse: fwiw I have around 30 versions that may or may not respond better to your shenanigans
reCurse: Try me
MSmits: and how would oyu adress it?
MSmits: that will be interesting reCurse
reCurse: Classified info
MSmits: oh ok, this is not common knowledge?
MSmits: my guess is that you are talking about some NN technique to remove bad flaws in the network that dont appear during training
MSmits: but can be exploited
MSmits: by a different kind of bot
reCurse: I'm talking about exploitation in the game theory sense
MSmits: so you would have to introduce some more randomness, the same kind you need to reduce overtraining?
MSmits: haha, dont worry, it's not like I can use the information. It's purely academic interest :)
struct: I thought fma did not work on CG cpu
struct: my bad
MSmits: you'll find this interesting reCurse
struct: Now I see it does
MSmits: parameter winrate
MSmits: added up: you + robo + agade
MSmits: see how sensitive it is
reCurse: Somehow not overfitting :thinking:
MSmits: well vs 3 NNs...
MSmits: its hard to overfit vs 3 separate bots isnt it?
reCurse: Not necessarily?
reCurse: You're talking about 3 bots that trained with more or less the same technique
MSmits: all versions of my bot had very different winrates vs your bots though, sometimes it won 0 vs Robo and 40 against you, sometimes the other way around
darkhorse64: > If I have correctlyHave you considered doing shallow
MSmits: but sure, i guess if you try enough combinations...
Astrobytes: broken ankle again darkhorse64? ;)
darkhorse64: Have you considered doing shallow minimax instead of evaluating nodes ?
AntiSquid: shallow ingrown toe nail
darkhorse64: Astrobytes : :stuck_out_tongue:
darkhorse64: MSmits: yes
MSmits: how does that work
darkhorse64: When you expand, you eval all child nodes, go one or two stages further à la minimax
MSmits: sounds very expensive though
Astrobytes: The tradeoff might be worth it
MSmits: would you then backprop 1 value or several?
Astrobytes: Only one?
MSmits: I mean... when i do an expansion now, I backprop each child
MSmits: added up
Astrobytes: Oh do you mean cumulative from whatever depth
darkhorse64: I have tried that with my poor eval: the bot was slightly weaker but the number of visited nodes grew to 8M first turn with depth 1 minimax
reCurse: You don't touch your tree
Astrobytes: ahhh the minimax is your eval score
MSmits: darkhorse64, you eval each child as well dont you? I mean without the minimax?
MSmits: and then back prop the total value?
MSmits: divided by the number of children?
MSmits: or am I doing that wrong
darkhorse64: Yes total value, and visits = number of nodes
MSmits: ok but with the minimax that is different?
Astrobytes: that's the score you backprop
Astrobytes: if I'm not mistaken
MSmits: yes but it's not 6 scores then?
MSmits: or is it no longer mcts?
MSmits: are you doing a best first kind of minimax?
Astrobytes: (I'm just guessing, I've only ever read about this)
darkhorse64: I have not played a lot with this idea mainly because my eval is not good. May be true minimax is better or just expansion with full backprop. Don't know but the number of explored nodes increase a lot and there may be something to gain from it
MSmits: hmm but why do the amount of explored nodes increase?
darkhorse64: You skip costly UCB
MSmits: right so, it's a full minimax tree?
MSmits: except it's best first?
darkhorse64: Actually, the UCB directs you to a group of nodes to explore not only at depth 0 but at 1 or 2
Zenoscave: I'm 1 Point behind marchet
MSmits: ahh ok, so you spend much more time in a branch
MSmits: less selection happening
darkhorse64: Yes. Once again, I have no proof that it works but it looks interestig
MSmits: what do you put on the nodes? Is this the minimax score? How does the minimax score turn into regular ucb score
MSmits: no i am very interested
Astrobytes: Guess some scaling is involved
Astrobytes: There's a few papers out there on it iirc, but don't take my virus-addled word for it :P
MSmits: yeah, I've seen some stuff like that, but not exactly that
darkhorse64: I put the scores at depth 1, the combined scores at depth 0, true back propagation. My reasoning is that I may explore further these nodes and I want to build my tree like MCTS does. That may be wrong
MSmits: yeah.. who knows
MSmits: it may depend on the game also
Astrobytes: Usually does it seems
darkhorse64: I have 5 MCTS bots, each has a different flavor
MSmits: mine are all different too
MSmits: do you use 0 and 1 for players or -1 and 1
MSmits: I like -1 and 1 for score multiplication but 0 and 1 for arrays
MSmits: win loss are always 1 and -1 for me
darkhorse64: 0, 1 for players, -1,0,1 for scores. I forgot to say that I use also eval scaled and clamped to 1
MSmits: ah same as I do in oware
MSmits: if this submit doesnt reach 1 I'll stop. reCurse is gonna submit a different version anyway. He's probably got several that beat my bot
MSmits: just wanna end 2 :)
darkhorse64: for my shallow expansion. My current bot is still > 0 = Win with early playout termination
MSmits: what is > 0 = ?
darkhorse64: Sorry. If eval > 0, count a win, etc ...
MSmits: ohh ok
MSmits: it improved mine when i changed
MSmits: but I really dont trust any parameter fit anymore
MSmits: I just test it vs the NN's and then it wins :P
MSmits: but reCurse is right, might just be an overfitting artefact...
darkhorse64: I remember you shortened rollouts when your eval improved. So it's not EPT anymore, shallow minimax is a natural extension on what you are doing
MSmits: it is kind of wasteful to go all the way down the tree and then not do very much :)
darkhorse64: Great! You push me and you lower the NNs. reCurse is going to think we are teaming
MSmits: you're welcome :)
MSmits: I was worried it would do badly against bots other than the nn's but it does ok
MSmits: I can keep it
Astrobytes: very nice MSmits
MSmits: sure, but if reCurse is right, it is also a bit cheesy maybe. I suspect that if any of you guys do a chain submit, it will leave reCurse mostly untouched and pull me back down
MSmits: his bot is more versatile
darkhorse64: For what I see, your winrate against me has decreased (from nearly 100 % to 80%)
Astrobytes: I'll see how many I can squeeze in tonight
MSmits: too litle statistics though darkhorse64
darkhorse64: Possibly but I see more wins from me in this submit than in the 10 previous ones
MSmits: you might be right
MSmits: this is also why it lost to my previous bot
MSmits: you can tell from the win % that i trained vs the NN's :P
MSmits: it's kind of cool that this is even possible though
MSmits: gonna try to orthogonalize some params
MSmits: try to make them less dependent on eachother, reduces the effect of overtraining, local minima and such maybe
MSmits: for example, kroo params grow with seedcount but I also have a seedcount param
MSmits: there should only be 1 of them growing
darkhorse64: Is that something you can really control ?
MSmits: well my kroo thingy is basically a growing periodic function
MSmits: I want to make it just periodic and put the growing into the seedcount
MSmits: eliminate 1 para
MSmits: so a fixed amplitude periodic function for the kroo, then a different param to give more score for higher seedcount
darkhorse64: growing with the #seeds in the kroo ?
AntiSquid: MSmits how come you second with that win rate?
MSmits: re curse had a lead
MSmits: not enough games in a single submit
MSmits: already did twice, pulled him down a little
AntiSquid: if you keep getting 70% you should get 1st no?
darkhorse64: periodic because a kroo that reaches the opponent side has more value
MSmits: but i dont chain submit unless it's deserved
AntiSquid: 70% win rate = deserved imo
MSmits: right now it's only deserved because I beat him after overtraining vs the nns. I do much worse than the nns versus the others
MSmits: so any of them submitting would pull me back down
MSmits: it's pointless
AntiSquid: pull you down how?
MSmits: by losing all battles vs re curse and winning 20% vs me
Astrobytes: Losing against the NN's with occasional wins agains Smits
Astrobytes: o beat me to it
MSmits: and another thing, i am probably just 70% vs this particular version of re curses bot
MSmits: and probably 30% vs another
MSmits: he can just submit a different one
AntiSquid: how do you know who even has a diff one?
MSmits: he just said he has 30 versions
MSmits: they take half an hour to train
darkhorse64: One for each core
MSmits: it takes days for me to fnd the right params to beat one of his versions :P
AntiSquid: 30 sounds redundant .
MSmits: if I want a permanent place on the nr 1, i could always go the book-route
MSmits: but I dont feel like it atm
AntiSquid: ugh just do it
MSmits: it takes a week to write the code
AntiSquid: it's the only way
MSmits: then another week or two to even get a good enough book
AntiSquid: it's your destiny MSmits, do it
MSmits: all the while I lose precious calc time I need to beat karliso at uttt :)
MSmits: I still win 45% :(
MSmits: moar games!
MSmits: Games played: 1,328,015 Nodes created: 3,068,857
AntiSquid: how come karliso always seem to wreck everyone else really badly?
MSmits: no idea, but he doesnt wreck me really badly
MSmits: he wrecks me a lil'
MSmits: he does have some weird opening book scheme that goes really deep in some branches and not at all in others
MSmits: I just do an exploration 2 wide search, not countering anyone
MSmits: I suppose i could just take his bot, note his starting sequence and counter that... but that's a temporary fix
MSmits: he can just run his own program for a day and end up with a new starting sequence
MSmits: nope... all i can do is... moar games!
MSmits: hey, be glad, I am not doing this on oware :p
reCurse: In a way you already are
MSmits: really.. I am just messing with 10 parameters
MSmits: you have thousands :P
Astrobytes: I think the end result is the same, or similar rather
MSmits: also the opening book thing is not overfitting if you just keep it to a single parameter
MSmits: and not counter anyone specific
MSmits: currently for uttt, that is what I do
MSmits: a NN is similar in that regard, it just tries to counter any "good" bot, not a specific bot.
MSmits: you're right that what i did with oware is a bit different. I was really just picking the only 3 bots that had any chance vs mine and use that as a reference. If there were more bots, I'd add them to the list. But yeah I guess that made it overfit =/
reCurse: Well I guess it's either I run a batch of random versions until I find an overfit too
reCurse: Or I finish my new training
reCurse: Or both
MSmits: I'm redoing my parameter setup to make it less vulnerable to overfitting. I really do prefer to have a bot that's strong overall
reCurse: As long as your baseline is batching against my bot and robo's it will overfit
MSmits: I added darkhorse now
MSmits: dunno what else to do
MSmits: its gonna be worse if i play against myself
MSmits: how should i fit params otherwise?
reCurse: With 10 parameters yeah
darkhorse64: I'm proud to be your sparring partner
MSmits: btw it's also Agade
MSmits: so 4 bots
reCurse: There's only 2 bots that matter
reCurse: And you know it
MSmits: I added him to reduce overfitting and because some of my versions lost a lot of games vs him
MSmits: the most typical result would be 70% winrate vs Agade, 40-50% vs you and somewhere between 40 and 70% vs Robo
MSmits: actually no, between 20 and 70... I overfit the most vs him I think
MSmits: but reCurse, there is no need to bother with submitting a new bot. you'll remain at 1
MSmits: at least, I wont spam this one more than the two submits I just did
MSmits: I will try another if I succeed in creating one with 2 less params in total that are also more orthogonal. it's an experiment.
MSmits: or maybe if i try darkhorse64's minimax idea
MadKnight: start what?
HungryBurger: I'm guessing Aguli is asking about how to use CodinGame to help him learn how to code.
Z_Doctor: I think you learn the basics of a language and you use this site to practice what you learned
HungryBurger: Indeed, that's about what I tell others.
HungryBurger: Ironically, I always credit CodinGame for "helping me learn Python" in the sense of "giving me so many opportunities to practice Python that I was able to teach myself".
miszu: Question for C++, why there is no core dump or error when your program has issue?