From CG community
Jump to navigation Jump to search

Default avatar.png ThayHaiDepZaiVaDeThuongQuaDiAk:

Default avatar.png YusufEmad04: fuck

jacek: oO

YCS-Venom: how i can store 109**987717 in string then output the first three characters in python?

jacek: str(109**987717)[:3]

eulerscheZahl: but memory will explode

eulerscheZahl: you could do something with logarithms

eulerscheZahl: 987717 * log10(109) to get a power of 10

eulerscheZahl: so this gives 2012400.7882664192

eulerscheZahl: you can make it smaller by dropping the integer part

jacek: is this... math? :scream:

eulerscheZahl: 10**0.7882664192 6.141386347439581

eulerscheZahl: so the first 3 digits: 614

Westicles: wolfram agrees. nicely done

jacek: you could just answer print(614)

YCS-Venom: i know and the time has exceeded the required amount

YCS-Venom: i tried that but it didn't give me something useful

jrke: i just calculated all possible sets of moves in dots and boxes

MSmits: wait what?

jrke: = 197450685722107402353682037275992488341277868034975337796656295094902858969771811440894224355027779366597957338237853638272334919686385621811850780464277094400000000000000000000000000

MSmits: you calculated the *number* of moves :P

jrke: all possible sets of moves yes

MSmits: and there's a lot of zeroes at the end so thats probably not exact

jrke: gamestates are less than that

MSmits: far far less

MSmits: this game has tons of transpositions, more than others, because lines are not player specific

jrke: i think factorial(112) are all possible sets of moves

MSmits: mmh isn't it 108 ?

MSmits: corners have 2 identical moves

jrke: i think 112 cause every match are 112 frames means 112 moves a game

MSmits: what i mean is, you're correct about the number of plies, but if two moves produce the exact same result, they're not really unique moves

MSmits: take the top left corner, it doesnt matter if you play the left line or the top line

MSmits: both give you 1 out of 4 lines for that box and nothing else

jrke: hmm

MSmits: this is only true for corners

MSmits: because there, they lack neighbors on 2 sides

MSmits: and there's also symmetry

MSmits: on the first turn you actually have only 14 moves

jrke: is there any way i can count total no. of unique gamestates?

jrke: 14 moves i didn't get that?

MSmits: you can rotate the board in 4 different ways

MSmits: and also mirror it

jrke: oh ye

MSmits: comes down to 8 rotation states

MSmits: 112/8 = 14

MSmits: works the same in uttt since thats also a square

MSmits: yavalath has 12 even, because it's hexagonal

jrke: hmm

jacek: hexagonal d&b :drooling_face:

MSmits: would be weird

MSmits: not sure if it would be interesting

jacek: everything is better with hexes

MSmits: takes a while to finish a box because it has 6 sides

MSmits: good rule of thumb yeah

jacek: thats universal law

MSmits: allright allright

jrke: in my d&b bot seed manipulation for generating randoms gives different ranks

MSmits: well that might be a statistical illusion

MSmits: dont trust the leaderboard

jrke: it means majority of bots are having randomness in it

MSmits: even deterministic minimax bots have some randomness

MSmits: even if they dont random on equal score

MSmits: the time limit is somewhat variable

MSmits: so sometimes you get 1 ply deeper in some turns

MSmits: which gives a different answer

jrke: hmm

MSmits: this not that common though

MSmits: so you would still mostly see the same games

MSmits: but if they random on equal score it can be a lot of random

MSmits: if the time limit wasn't variable then even mcts bots could be deterministic if their random seed is always the same

jrke: what you like random bots or stable bots

MSmits: but mcts bots are more affected by the time limit being variable than minimax bots are, because they dont add an entire depth level at once, but more gradually deepen a tree, so 1 added game can make the difference

MSmits: jrke it depends, if you're not the best bot and there are a lot of opponents, a non-random bot is fine

MSmits: but if the field is small or your at the top, you better add some random

MSmits: you're

MSmits: it's the whole reason jacek is still number 1 oware

jrke: hmm

jrke: i think crazyremi is having Ml based bot in D&B

MSmits: are you saying that just because he's nr 1?

MSmits: ML is not an i-win button. In some games it underperforms, many in fact

jrke: no look at points difference

MSmits: yes, look at my point difference on othello and yavalath. I dont use ML

MSmits: it's just a nice opening book is all

jrke: hmm

MSmits: of course thats not what remi uses

MSmits: but he may just have better heuristics

MSmits: 1 heuristic can be a 10 point rating difference

MSmits: technically

MSmits: usually if its that strong, others will have found it

MSmits: but if you have one, noone else knows...

jrke: yup

MSmits: you could do some research on D&B, see if there's something there

MSmits: i know a ton of heuristics, which is how i wrote my solver. But my heuristics are only late game

MSmits: not sure how to do the early game

MSmits: might be that the fact that i can solve at ply 50, doesnt help me

MSmits: because something smart is happening before that

jrke: for winning endgame you need to have better position in game opening turns

MSmits: yeah but what is a better position?

MSmits: both players uses the same board

jacek: i think remi said he uses just some heuristic, even not search

MSmits: so it's just who's turn it is that decides who has the better position

MSmits: yeah i think he uses nimbers

MSmits: nim theory

MSmits: counting chains

MSmits: do you do something with this"?


jrke: mine is just depth 3 minimax i think a bit optim. can give me depth 4 minimax

MSmits: yes but how do you score?

MSmits: an early game

MSmits: eval score

MSmits: it's nice that you know what the board looks like in 3-4 plies, but if you dont know whats good and whats bad...

MSmits: jacek yeah i saw that, makes total sense to me that azero doesnt work

jrke: my minimax is bit different after each move i see if there is any chain is there i play it and revert last 2 moves of chain playing

MSmits: it seems really hard to figure out what the board value is

MSmits: and both players use the same board

jrke: thats the problem i want early game eval

jacek: opening book *.*

MSmits: ah ok jrke, so you're basically where i am, but without a solver

MSmits: i was hoping you had some secrets i could steal :P

MSmits: maybe you do and you dont realize it

MSmits: your rank is pretty good after all

MSmits: jacek i doubt opening book works here. I can see it working against deterministic bots only

MSmits: too much branching otherwise

jrke: its my favourite game i will rewrite it

MSmits: i like it too, but i am really dreading putting my solver in and finishing the bot. The solver alone is 1500 lines :(

MSmits: i was going to do random moves basically until endgame hits, but an endgame state has many possible situations

MSmits: you can be busy finishing a chain, having to choose between giving up control or not etc.

MSmits: i need a lot of conditionals and link them up with the solver

MSmits: and that's with a really complex bitboard

jrke: i will say all the best

jrke: some work gtg bye

MSmits: allright have a good day!

AntiSquid: but not dreading 1500 paragraphs of chat ? :thinking:

MSmits: no, that comes natural

AntiSquid: so how much did you check different solver outcomes?

MSmits: what do you mean exactly by check?

MSmits: check if they are correct, by hand?

AntiSquid: test / brutal test / submit spam test any, all of the above

AntiSquid: efficiency

MSmits: ahh I did not do much verification, but if you mean just generating random endgame states, i did that millions of times

MSmits: the average endgame state solves in 1 ms, about 1 in a millon states cant be solved in 100 ms

MSmits: an endgame state is defined as a state where you cant place a line without giving away a box

MSmits: so thats halfway through the game

MSmits: some states are ridiculously complex, i dont know if they actually occur since my states are generated randomly, not intelligently

MSmits: think states with 12 three and four way splits

AntiSquid: no heuristic shortcuts to get quicker to the best move for the 1 in mil states?

MSmits: there are a huge amount of heuristis in my solver

MSmits: so i already do this

MSmits: i really scoured those D&B papers for them

MSmits: used a lot

AntiSquid: :D

MSmits: mind you, halfway through the game there are like 56 lines left to play

MSmits: what other game can be solved at that point

MSmits: it's 56! possible games left to play

MSmits: (factorial)

AntiSquid: idk, how similar is paper soccer to dots and boxes? i didn't try paper soccer at all

MSmits: it seems similar at first glance, but i think it might be very different

MSmits: i havent tried either

MSmits: i know you place lines and you get to bounce multiple moves on your turn

MSmits: but you dont get points for finishing boxes, you just win, or not

MSmits: by scoring

MSmits: now i want to try and finish D&B and i was just gonna rewrite my uttt

MSmits: damnit

eulerscheZahl: there are more games than uttt and yavalath ;)

MSmits: yeah, too many :(

AntiSquid: just seen these news:

MSmits: lol

eulerscheZahl: i saw a youtube video about that village


MSmits: i am surprised its still on YT

eulerscheZahl: me too

AntiSquid: why? people swear a lot on youtube, nothing uncommon

MSmits: did you watch the whole video?

MSmits: i mean i dont say it should be off YT, i just think they might have

eulerscheZahl: yeah, because YT

eulerscheZahl: they like to delete when in doubt

MSmits: right

eulerscheZahl: i also don't always understand US logic. weapons? fine. nudity? no way, that's dangerous for the children

MSmits: it's their religious beliefs mostly I think

eulerscheZahl: do you know Kenny vs Spenny?

MSmits: nope

MSmits: sounds funny

AntiSquid: reminds me of borat making a documentary about a village in "azerbaijan" which was actually in romania and talking shit about the villagers without them knowing, later they tried sueing, wonder if this guy got sued

eulerscheZahl: 2 guys having competitions, sometimes funny, sometimes less

MSmits: did they kill kenny?/

eulerscheZahl: on episode: who can piss off more people

eulerscheZahl: kenny rented and airplane and added a banner to it "Jusus sucks"

eulerscheZahl: Jesus*

MSmits: depends if they have a sense of humor AntiSquid, he's only making fun of the name itself

eulerscheZahl: franatic people calling the radio channel, he won that round

MSmits: lol


jacek: paper soccer is not similar to d&b. its more awesome

AntiSquid: i think youtube is more set on deleting anything that doesn't fit a specific point of view atm, covid, politicallly related stuff that doesn't support specific views and popular channels regarding occult conspiracy theories

AntiSquid: MSmits actually he lied to them, paid each 3$ or something and then was talking some serious shit about them, but didn't tell them what he was saying, lied instead, later they found out when it wa son TV

AntiSquid: he told them it was a documentary lol

AntiSquid: somewhat like the guy in the austrian village

MSmits: ahh ok

jacek: stupid Y, n-tuples prone to overfitting

AntiSquid: what are you working on jacek?

AntiSquid: which game i mean

jacek: yavalath

AntiSquid: ok now i am curious what your model looks like :P

jacek: just all possible 4-rows

**AntiSquid confused

jacek: hm?

AntiSquid: i just don't know what you mean by that, is every row in the 4 directions an input?

MSmits: every way you can win with 4 in a row

jacek: n-tuple is pattern system

MSmits: so 4 hexes lined up

jacek: i extracts those patterns

AntiSquid: so you only give it the 4 in a row it can complete?

MSmits: no, on eval, he checks every possible 4 pattern and sees whats on there (empty, p1, p2) each of these possibilities gets a score

MSmits: everything is added up

jacek: there are 102 row-of-4patterns. each pattern has its weight like x..x is something else than xx.x. and i sum them all

AntiSquid: ah you do scoring for each possible move then

MSmits: i am guessing it is an incremental eval

MSmits: so when he plays a move, he updates the eval

MSmits: only refreshes the patterns affected by the move

jacek: mhm

MSmits: thats what I would do anyway...

MSmits: dont tell me you redo the entire eval?

AntiSquid: was thinking of making it give you the output for the best move

MSmits: that would be slow

jacek: i use nn or n-tuple for eval only

MSmits: which makes sense but I also understand why azero includes the action

MSmits: I think it is because you can reach the same state by different routes, it might converge better if you include the action

jacek: they can use gpu/tpu to output entire move set. justifies the latency between cpu-gpu

MSmits: well there is that

MSmits: btw, i did a minimax with exactly the same patterns and I was slightly weaker than your pony is now

MSmits: i just picked the params by hand

MSmits: I think i had 6

MSmits: like if x00x being better than 0xx0

MSmits: or something

AntiSquid: alt spotted :x

MSmits: i never realized i could instead have thousands of params here

MSmits: you didnt know derjack was jaceks alt?

AntiSquid: i didn't even know it existed :P

jacek: well there is also position

MSmits: yeah there is jacek, but i didnt have the tools to take that into account

MSmits: i figured with enough depth it would be ok

MSmits: maybe if i did jacekmax it would be way stronger

MSmits: but then the incremental eval wouldnt work right =/

jacek: :thinking:

MSmits: jacekmax is nice here because there's so much branching and so many crap moves

derjack: eeyup

derjack: in forced move branching is effectively only 1

MSmits: this is a funny game

MSmits: look near the end, it seems like i am letting him win by not blocking

MSmits: but he had a trap ready to go for quite a few turns and he didnt tak eit

MSmits: (at 6 5)

MSmits: so my solver just says fk it

derjack: hes taunting you

MSmits: seems so

derjack: or because there is 4-in-row and 3-in-row simulatouensnsnsly?

MSmits: oh hmm, yeah i missed that

derjack: maybe he has bug

MSmits: both bot bots say it's solved, so must be some other way

MSmits: nah i doubt that

MSmits: no i didnt miss it

MSmits: 6 5 works

MSmits: he doesnt have 3 in a row there, i do

MSmits: i guess both moves led to a win for him and his bot doesnt care about early or late wins

MSmits: another loss by me. Detected 20 plies before the end lol


geppoz: this reminds me a game i did long time ago


geppoz: only similar for hex btw, that is a solitaire puzzle ;)

geppoz: at increasing difficulty

AntiSquid: wtf is that link ?

geppoz: an applet

geppoz: i did time ago

derjack: is it legit

geppoz: a game

AntiSquid: i mean why does it ask to download stuff

geppoz: it is an applet

geppoz: no more supported by browsers

geppoz: so now you need to use locally, but nvm

geppoz: ignore it

derjack: like flash

AntiSquid: ok that explains it, typical java

geppoz: legit to not trust me ;)

geppoz: someday I will convert it in JS

derjack: :scream:

geppoz: that was a very bad moment when browsers removed applet support :(

AntiSquid: how long will JS fail to get the recognition it deserves ? :(

MSmits: how long will JS fail?

AntiSquid: you're using it right now MSmits ! think that's a success

MSmits: :grin:

MSmits: I'm just messing with you

AntiSquid: even Automaton2000 noticed that

Automaton2000: the last one is the current state

MSmits: this is how Automaton2000 says he won

Automaton2000: to ensure that you don't need to put in a lot of experience

AntiSquid: this is how he says: Big Brother Automaton2000 is watching you

Automaton2000: so you have to do the same for all languages

jacek: AutomatonNN eh

AutomatonNN: what do you mean by "if you can read it in the list of the context" ?

MSmits: AutomatonNN say something about euler

AutomatonNN: else i can get the points for coding

jacek: AutomatonNN who is noob

AutomatonNN: gg

AntiSquid: you get social points for saying euler, he'll like you more AutomatonNN

AutomatonNN: where is the problem?

Default avatar.png SI_PROVA: yo guys how can i optimize the fking "horse-racing duals" puzzle

derjack: pony racing?

Default avatar.png SI_PROVA: the easy puzzle

eulerscheZahl: click the hints on the left

derjack: timeout?

Default avatar.png SI_PROVA: yeah time out

Default avatar.png SI_PROVA: i've got it in the hints

Default avatar.png SI_PROVA: thx

ParticleBeam: I hate working with Laptop keyboards

eulerscheZahl: same

eulerscheZahl: and keyboards that have a touchpad below

eulerscheZahl: you type something and accidentally click

eulerscheZahl: and continue writing at another location

jacek: hmm

jacek: fish is broken

eulerscheZahl: epic win :P

eulerscheZahl: give him some chips

MSmits: from what i understand he uses null window

MSmits: sometimes it causes problems

MSmits: wait maybe it was something else

MSmits: anyways its some pruning mechanic that sometimes misses something

MSmits: oh its null move

MSmits: not null window, thats something else

derjack: this eval sometimes sucks

MSmits: seems to detect win/loss very late

MSmits: assuming darkhorse did this correctly

derjack: or some bug with solver hmm

MSmits: could be

darkhorse64: I think my solver is correct but the depth is the closest to victory I have found at the moment not necessarily the best

MSmits: yeah i didnt find any errors in your solver

darkhorse64: In this game, I am proud of 65 prepared the devilish trap 42

MSmits: yeah its fine

darkhorse64: *preparing*

MSmits: well played. I think if i spam submit, your bot ranks quite high because its very random

MSmits: i cant book very well against it

MSmits: and it does well in late game

MSmits: i win most games, but not all

darkhorse64: My book excludes the seven center starting moves. I should try to add an eval or something that biaises towards 4 length triangles

darkhorse64: Astrobytes: you ask me (and others) a question yesterday but, due to lack of contest or not enough history in chat, I failed to understand

MSmits: nice typo, lack of contest :)

darkhorse64: :slight_smile:

Astrobytes: oh, I believe it was just for some advice for Zenoscave, he was looking into MCTS solver and whatnot, MSmits was there


eulerscheZahl: there's your contest

Astrobytes: lol

jacek: another victim of the mcts solver paper?

Astrobytes: I warned him against reading it :)

MSmits: well the idea is good, it's just the pseudocode that's crap

Astrobytes: yeah

eulerscheZahl: theoretic scientists writing code...

eulerscheZahl: once a prof told a story about someone from research implementing an algo

eulerscheZahl: was way faster than expected

eulerscheZahl: turned out he added a bug that made it faster (but still correct)

eulerscheZahl: something about minimum set cover iirc

eulerscheZahl: vertex cover

MSmits: so the moral of the story is that researchers only produce fast code when they're bugging it?

Astrobytes: researchers introduce helpful bugs

eulerscheZahl: they can't even follow the simplest of instructions :D

Astrobytes: "After 12 minutes we reached depth 5..."

MSmits: ok so == has priority over | ?

Astrobytes: lots of papers like that

MSmits: damn bit ops, i should know this by now

jacek: here you go: ( )

MSmits: thanks jacek, lifesaver

eulerscheZahl: jacek coding clojure on the chat?

jacek: :(){ :|:& };:

eulerscheZahl: won't paste that in my terminal

eulerscheZahl: not again

jacek: how about CG ide

jacek: hmm

MSmits: what happens

eulerscheZahl: fork bomb

eulerscheZahl: spawns new processes

MSmits: is that bash or something?

eulerscheZahl: mouse will get laggy at first

eulerscheZahl: then nothing responds anymore

eulerscheZahl: yes, bash

MSmits: evil

eulerscheZahl: you'll have to hard-reset your computer

MSmits: but it's so short

MSmits: i remember trying to create a process from a windows form, was a lot more work

eulerscheZahl: i bought that one :D

MSmits: i bet virus creators love this

MSmits: lol

struct: time to reuse connect 4 ais

jacek: :nerd:

darkhorse64: 4 hours coding to improve my score at A*C by 30 points :disappointed:

MSmits: :(

reCurse: Life in a nutshell

darkhorse64: Worse, I hardly beat a previous version which computed wrong scores. Once the bug was fixed, it performed worse ... :angry:

MSmits: the worst bugs are also features

reCurse: "Once the bug was fixed it performed worse" is like the chuck norris equivalent for multiplayer

darkhorse64: :nerd:

eulerscheZahl: confession: i didn't like the chuck norris puzzle

reCurse: Since there's so much momentum for correct representation in emojis, we should have different glasses shape

eulerscheZahl: we have :beer: and :beers:

eulerscheZahl: :wine_glass:

reCurse: What about liquors uh? Second class beverages is that it?

eulerscheZahl: all the same for me, i don't drink alcohol

Astrobytes: yeah yeah, we all know you drink a pint of schnapps at 7pm and that's why you go to bed at 8 :P

reCurse: First logical explanation I hear

eulerscheZahl: first of all it's around 8:30 - 9:00pm

Astrobytes: Early to bed, early to rise, makes a man healthy, wealthy and eulerscheZahl

reCurse: Old?

Astrobytes: :older_man:

darkhorse64: Reading the PMs, I have tried to classify cells according various characteristics to exclude some from search and to start searching first with edges cells, then cells with arrows in the neighborhood, trying to refine my solutions while enlarging the search space. It's an improvement but I am still far from the 11k mark

reCurse: Have you tried hardcoding the solutions?

eulerscheZahl: the answer is offline solving

eulerscheZahl: and reCurse wins by type speed

reCurse: Flexing that wpm

darkhorse64: Some did it without. Why not me ?

MSmits: reCurse dont answer, that's rhetorical!

reCurse: By the same logic, some did it with, why not you?

MSmits: sigh, i knew it :P

Astrobytes: MSmits :grin:

reCurse: The day I stop engaging in those is the day euler goes to bed at midnight.

MSmits: so just after hell freezes over

Astrobytes: Approximately.

darkhorse64: I prefer to devise a better algorithm

reCurse: Have fun with the spam submit then :P

MSmits: did you do SA?

darkhorse64: LAHC and GA, LAHC fares better maybe I should start tweaking params

MSmits: tourist did SA

Astrobytes: with some tricks

MSmits: maybe SA works better in the short calc time available

darkhorse64: LAHC is a kind of SA.

MSmits: ah ok

darkhorse64: I "borrowed" it from Marchete

MSmits: sometimes it helps to prune the searchspace

MSmits: so keep the search the same, but disallow some changes

Astrobytes: here was Smits thinking it was Large-Ass Hadron Collider

MSmits: makes sense to me

Astrobytes: Marchet e used it for... NS right?

eulerscheZahl: yes

Astrobytes: I'd never heard of it until then

darkhorse64: I use some eZ tricks, if matation = previous state, remove the arrow. Nice improvement

MSmits: hakuna matation

Astrobytes: matations are mutations that only happen in the morning

MSmits: oh ok

darkhorse64: Other people are a treasure trove for knowledge, that's what I like here. This is my :heart_eyes: minute

eulerscheZahl: matation is mutation applied to mutation

reCurse: Genetic mistake

Astrobytes: matin + mutation = matation

reCurse: Hmm... GA should stand for genetic anomaly now that I think of it

darkhorse64: morning coding = crash dump

Astrobytes: hahaha

eulerscheZahl: in the morning i write the best code

reCurse: Your morning is what most people consider their afternoon

darkhorse64: in the morning, I sleep

Astrobytes: :D

darkhorse64: early in the morning, mourning

darkhorse64: OK, time to write that NS solver

Skynse: I write the best code with red eyes

Skynse: That is, when I desire the bed more than anything

Skynse: dang, good thing there are always moderators online

jacek: oO

Skynse: Cause some weird scenario popped up in my head where I needed moderators to ban some dude lol

Skynse: but eh, I was just daydreaming again

jacek: mk's avatar miniature looks like karen's hair

MadKnight: lol why

AntiSquid: strange fetish

Default avatar.png famouswithoutname: Hi everyone

jacek: MSmits do you how much stronger is your bot in bandas with endtables vs without endtables?

MSmits: in terms of game rank it's 2-3 instead of 8

MSmits: leaderboard rank i mean

Default avatar.png Knee-Gears: i droped from 1 to 24k :(

jacek: hmm, i have only slight increase winrate if i use more time on board size <= 16 (simulating endtabled)

MSmits: wait, how do you give it extra time specifically for that board size?

jacek: i mean locally

MSmits: yes i know

MSmits: but how do you give it extra time for that board size

MSmits: how does this work exactly

jacek: i dont use cg arena and whatnot

MSmits: i know, but i mean, the mechanics of the mcts

MSmits: the tree etc.

MSmits: how does the small board size get extra time

jacek: auto move = cpu2.getBestMove(game->board.getSize() <= 16 ? 2000 * 1000L : 10 * 1000L);

MSmits: allright, so you wait till the game is to a stage where the board is small and then it gets extra time?

jacek: yes

MSmits: that doesnt work

MSmits: we dont use the endgame table when the game gets to that stage

MSmits: we use it from turn 1

MSmits: when the board is 8x8

MSmits: remember, there are rollouts

jacek: ah, in mcts simulation?

MSmits: yes

MSmits: thats why my games are often solved at around ply 7-15 way before it gets small even

jacek: meh

MSmits: when i expand into a node with a board of that size, there isnt even any rollout

MSmits: thats when the solver kicks in

jacek: i wonder how would i simulate that. i would need fixed rollouts instead of time

jacek: to see how much benefits

MSmits: it's pretty safe to assume it helps a lot

MSmits: karliso nagrarok and I all use it

MSmits: remember that many games are decided on turn 1 because they are unbalanced

MSmits: so to get this much rating difference, means we do really well on the more balanced games

MSmits: helps more than in oware anyway

MSmits: in oware it's debatable

MSmits: I like them which is why i dont mind that it doesnt help me much

Default avatar.png Toby75: yeah

Default avatar.png Knee-Gears: i have ebola

MSmits: quite an achievement

MSmits: where did you order your electronic bolas?

Default avatar.png Knee-Gears: from 'niger' in africa

Default avatar.png Knee-Gears: lol smithy how are you 29th world wide xDD

jacek: oO

MSmits: lack of a social life

Default avatar.png Knee-Gears: lamooo

Default avatar.png Knee-Gears: lmaooo*

Default avatar.png Toby75: how old are you, if I may ask?

Default avatar.png Toby75: Because your profile says that you are a teacher

MSmits: me or the troll?

Default avatar.png Toby75: you Smits

MSmits: ah, I'm almost 40

Default avatar.png Knee-Gears: im in a discord call with toby

Default avatar.png Toby75: nice

Default avatar.png Toby75: I'm very young, and I'm looking forward to become a good coder

Default avatar.png Knee-Gears: same

Default avatar.png Toby75: But I don't know how to practise stuff like algorithm

MSmits: first learn the basics of coding

MSmits: loops, functions, variables etc.

Default avatar.png Toby75: I am aware of them

MSmits: then do some easy puzzles, maybe use python

MSmits: work your way up from there

Default avatar.png Toby75: So you recommend python for beginners?

MSmits: definitely

Default avatar.png Knee-Gears: damm why not js

Default avatar.png Toby75: But I heared that python is so diffrent to other languages...

MSmits: js can work too

Default avatar.png Knee-Gears: ok

MSmits: all languages are different, some share similarities

Default avatar.png Knee-Gears: *like java and js*

MSmits: python is very popular so the face that there isnt another language that is similar to it, is not a problem

MSmits: fact

Default avatar.png Toby75: And how do you recommend practising algorithm?

Default avatar.png Toby75: Just with CodinGame?

jacek: on CG, puzzles

MSmits: that's a really vague question though, an algorithm is just a series of instructions for a computer

MSmits: so any program will be using an algorithm

Default avatar.png Toby75: Yeah, but I mean how to create good ones

jacek: though for very basic stuff you would need some tutorial outside

Default avatar.png Knee-Gears: aight imma heat out by smith, hope to see you on some other day

Default avatar.png Toby75: or how to create own, efficient ones to solve certain problems

Default avatar.png Knee-Gears: bye*

MSmits: algorithms aren't necessary good or bad, they need to fit with the situation

MSmits: if you see a puzzle and you come up with a solution, you just wrote an algorithm

MSmits: so just practice

MSmits: Knee-Gears bye

Default avatar.png Toby75: okay thank you

Default avatar.png Toby75: you are amazing

MSmits: thanks

struct: MSmits I have a question sinse you are used to using windows forms

MSmits: well.... dont call me an expert. I got it to work and i do the same thing every time now

struct: I want to run c++ program, I know how to connect it with C#

MSmits: ah yes, sec

MSmits: that i know

jacek: C#++?

struct: No, not my question

struct: One sec

MSmits: well, i dont know the DLL stuff, you can create a dynamic link library. I just know how to start and kill a process. Is that good enough?

struct: Well I would like to pass all precomputed states to the C# program

struct: So I can see them

Default avatar.png Toby75: How long does it take you to solve very hard coding challenges?

struct: The number can be quite big,

struct: How should I pass them

MSmits: mmh you might want to just serialize it

MSmits: write to file and read

struct: Ok, I guess I can do that

MSmits: Toby75 too long, just do what is fun

struct: This STC is giving me too much work

Default avatar.png Toby75: My problem is that in the most cases I just don't understand the given examples

jacek: ok, having <= 12 'endtables' increased winterate to around 55-60%

jacek: for example?

MSmits: i'll share my arena class for windows form

Default avatar.png Toby75: Maybe because my english language skill is lacking

MSmits: might be Toby75

MSmits: struct:


Default avatar.png Toby75: is it ok to google some stuff when solving a problem?

MSmits: thats my UTTT arena class for C#

struct: Thanks

MSmits: I call this from a window form to create two processes and handle input and output

MSmits: i dont need a referee since i use the bots solver output to tell me the game i sover

jacek: who uses { on new line :unamused:

MSmits: me :P

struct: Yeah, I probably write the referee

MSmits: I dont know, i like the symmetry

Default avatar.png Toby75: It's easier to recognize the brackets

MSmits: yeah

MSmits: thanks Toby75 thats indeed why

MSmits: but i know almost everyone doesnt do it

MSmits: I also write C++ in camelcase

Default avatar.png Toby75: But I would never ever be able to use this in my code, just because of the effort it would cost me to format it like this :sweat_smile:

MSmits: because of my C# roots

struct: I blame C# for that

MSmits: Toby75 it's important to have a style and be consistent

MSmits: helps read your own code

Default avatar.png Toby75: definitely

MSmits: of course its best to code in industry standard, because that helps you read others

MSmits: but that ship has sailed for me :P

Default avatar.png Toby75: but what IS industry standard?

MSmits: for c++ it is

MSmits: this_variable

MSmits: instead of thisVariable

MSmits: and functions with lower case

MSmits: and I use uppercase etc.

Default avatar.png Toby75: aaahh

jacek: for python you have pep8 guidelines, for others other things

MSmits: yeah i try to use the guidelines for python since i learned it late and i have to set an example for my students

jacek: C# is just java with Capital Method Names

jacek: pycharm will nicely warn you about style

struct: I need to write this "arena" because I cant find the bug

struct: Im reusing states, but sometimes it bugs :/

MSmits: earlier today I found a solved state in my meta mcts i had to unsolve because it was solved wrong =/

MSmits: robo won a game through this state

RoboStac: dammit, he's found my backdoor

MSmits: lol

MSmits: what did you fix about your bot btw Robo?

MSmits: or is it just having random success again?

RoboStac: just retrained with the improvements I made for othello

MSmits: ah ok

RoboStac: I realised I'd never got round to redoing yava

jacek: what game you need to resolve?

MSmits: dont have the link

jacek: oware, Y, other?

MSmits: Y

MSmits: I think I copied the library from an old version of my meta mcts that still had a bug

MSmits: either that or it's just a really rare bug

RoboStac: the main imrpovement was allowing a much bigger network via base65536

jacek: :scream:

MSmits: it's not exactly that is it? I still havent gotten past the unicode invalid characters

RoboStac: I'm usin gbase65536 exactly and just accepting that I can't quite get to 200kb

MSmits: does it skip some characters or?

MSmits: you shared it with us before

MSmits: i have it saved somewhere

jacek: im using int16 from 0 to 50000, so each unicode is 1 character

RoboStac: yeah, it only uses valid characters, but some count as 2

jacek: with a little tiny loss of precision

RoboStac: my bots only 175kb but 99k chars

MSmits: ah, i have two ranges, there's an invalid range of characters about 2k wide somewhere above 50k

jacek: i sent pm to agad how i do this

jacek: might as well share it

MSmits: agad needed it for oware NN i guess

Default avatar.png famouswithoutname: can somebody help with "Dungeon and Maps" task? Python

jacek: well, its pathfinding, no?

MSmits: I'm still waiting for my bots to need compression. Biggest books are 22kb, but they all use integer numbers for moves and are space separated, so can be compressed by a factor of 20 or so

Default avatar.png famouswithoutname: Yes

struct: This biggest book is for which multi?

RoboStac: I guess I should retrain bandas too at some point

MSmits: RoboStac, add endgametables if you've got around 20-30 kb to spare

MSmits: thats how big mine are

MSmits: the next step is > 200 kb

MSmits: so thats about as big as they can get

RoboStac: in theory I should do if I convert it from base85

MSmits: I expect that if you do all that you'll be nr 1

MSmits: endgame tables are quite a challenge, take a day for that one

MSmits: I;m saying a day cuz it took me a week

struct: so a month for me

MSmits: when creating games you're quite fast though

struct: for simple stuff I can do it fast yeah

struct: If it doesnt require stuff I dont already know

MSmits: ah ok

MSmits: I suppose it's often the thinking that takes the most time

struct: for me its learning

struct: I struggle a lot reading papers

jacek: like the one about mcts solver?

MSmits: there havent been many games where i actually had to go for the papers

MSmits: mostly the mcts stuff and D&B

MSmits: I struggle too, i think you're supposed to struggle

MSmits: it's meant for experts in the field

struct: Even sometimes for algorithms

struct: reading is not enough for me

MSmits: well i have to read some of that stuff 10 times before i get it

MSmits: and sometimes it lacks context and i never get it

RoboStac: so my bots learnt to beat smits but now has a fish problem :(

jacek: sounds fishy

jacek: exploit the fact fish has some bug with null move heuristic

jacek: your bot is alphazero style with conv net?

RoboStac: yeah

MSmits: your bot learned to beat the current version of my book :P

MSmits: busy fixing a lost game vs darkhorse, will get to it tomorrow

MSmits: ohh these are really short games

MSmits: 13-14 plies, very cool

MSmits: let me switch to that

jacek: so may games, so little cores

jacek: good winter is coming

MSmits: yeah, i played a move my meta mcts already solved as a loss.

MSmits: i just hadnt put it in my book yet I suppose

jacek: such a disgrace

karliso: MSmits Is it not better to overtake (2, 3)?

MSmits: you mean steal? Nah, not for me anyway

MSmits: it's pretty balanced

MSmits: so i can win games when i take and when i dont take

MSmits: but i have somewhat better results not stealing

RoboStac: I probably should fix my starts at some point

RoboStac: still hardcoded first move as p1 + always steal as p2

MSmits: i have had starts with 1 0 and 2 1 as well, they are fine, tric trac does this

karliso: Are there draws in Y?

RoboStac: yes, but rare

MSmits: yes, but rare

MSmits: nice


karliso: You are on the same wavelength

MSmits: apparently :)

MSmits: the game might be solved as a draw

MSmits: to not be solved as a draw, every starting move needs to have a winner

RoboStac: though saying rare I did draw vs tic 3 times out of 9

RoboStac: on my last submit

MSmits: its funny, every game i lose vs you is identical

MSmits: oh no i found a different one

MSmits: loss in the same place though

MSmits: just different losing move :P

RoboStac: theres 3 - lengths 11/12/14

RoboStac: though I guess 11/12 are the same but reversed players

MSmits: could be, also once my bot solved as a loss, it doesnt really care anymore, might take longer sometimes

MSmits: never more than 1-2 plies, depends if it is a trap or a double 4 in a row

RoboStac: and the 14 is basically the same except my bot makes you play one extra forced move

MSmits: ah yeah

karliso: Wouldn't Y get boring with large opening books?

MSmits: they aren't exactly large though, mine isnt

karliso: exactly

MSmits: 1308 moves

MSmits: of course rotated for symmetry, so around 15k in that case

MSmits: i dont think it's boring because I like finding the moves

karliso: What if someone else approached openings like you and you both pushed each other.

karliso: Would there still be middlegame left?

MSmits: hasnt happened, would be interesting

MSmits: we almost had something like that in othello with oldjohn

MSmits: but he stopped

MSmits: the main thing is that my bot is really strong late game

MSmits: so i dont have to worry past a certain point

RoboStac: I imagine mines quite weak due to low sims + not much training data

MSmits: it's strong early game though

MSmits: and thats where most wins happen

Moogs: umu

MSmits: my meta mcts just said i need to break your triangle robo

jacek: the shell has spoken!

MSmits: on my 3rd hex, i somehow decide to place it outside of the triangle area

MSmits: the mcts has a serious depth problem early game

MSmits: most of my moves are triangle fixes, often vs dafish

MSmits: but karliso as for whether it is boring. I like the game itself because of the tactics, traps etc. So I like watching the meta mcts solve nodes and come up with counters. Thats why its not boring

MSmits: seeing robo wreck me 10x in a row is exciting :P

jacek: it just sucks seeing bot missing obvious traps and game ending in 11-12 plies

MSmits: jacek do you think maybe a ntuple bot isnt enough here? I think NN might fare better

MSmits: traps have odd shapes

MSmits: you have to have a relation between two 4-patterns to spot it

MSmits: NN layers do this

jacek: maybe. or maybe i need other approach to create my training data. the winrate can jump 45%-75% against baseline

MSmits: you should check how many different games it plays though

MSmits: it might be deterministic and then the winrate doesnt mean much

jacek: yeah it was pretty deterministic so i added more random

jacek: but still its jumping

MSmits: hm ok

MSmits: I'm not sure if the starting move might not have a huge effect on the flow of the game

MSmits: a center start is very different from a corner start it seems

struct: Is there any move that is always a loss?

MSmits: you mean, not considering the steal?

struct: yeah

MSmits: 0 0 and 2 0 are very weak

MSmits: I havent solved them though

MSmits: the +1 hex advantage of player 1 is quite substantial so it compensates the bad start a bit

MSmits: for example, the 4 4 start is a solved game

MSmits: but player 1 does 0 0

MSmits: and then player 2 does 4 4

MSmits: not solved

MSmits: even though the 0 0 is entirely out of range

MSmits: so it seems as if p2 an use the same tactics to win

MSmits: but at some point that corner hex will generate some threat that prevents the 4 4 win

MSmits: best counter to 0 0 is 3 3

MSmits: blocks 4 in a row coming from the corner

Default avatar.png RyanGilbert: I am streaming, anybody want to watch or do a private clash?

Default avatar.png RyanGilbert:

Default avatar.png RyanGilbert:

Astrobytes: RyanGilbert: it's already in #clash, up the top on stream notifications, and discord. Please desist.

Default avatar.png RyanGilbert: OK, my apologies

MSmits: it's ok to say here you're streaming, if you keep it to one message right? Just no posting clash links?

Astrobytes: yeah totally

Astrobytes: no worries RyanGilbert

Default avatar.png RyanGilbert: OK cool

jacek: happy Princess Luna Day

Default avatar.png RyanGilbert: Streaming again to those who were waiting

Astrobytes: RyanGilbert your stream notification has disappeared

Default avatar.png RyanGilbert: On discord?

Astrobytes: on here and discord it seems

Default avatar.png RyanGilbert: hm ok

Default avatar.png RyanGilbert: give it a few minutes probably

Astrobytes: last one is from about 15 minutes ago

Astrobytes: yeah, probably just twitch being twitch

Default avatar.png RyanGilbert: any reason for the lag?

Default avatar.png FirstInLastOut: so i'm not alone lagging ..

Default avatar.png TM_Tsukiru: Here too

Default avatar.png RyanGilbert: yeah

IamFish: Usually lags around this time. Should get better in like 30 min in my experience

Default avatar.png RyanGilbert: ok

Default avatar.png NeoXX: hi

Default avatar.png NeoXX: henlLOOO

Default avatar.png FirstInLastOut: hello

Default avatar.png NeoXX: WHAT IS THIS?!?!?

Default avatar.png NeoXX: WHOAAA

Default avatar.png NeoXX: SOMEONE SAID SOMETHING

Default avatar.png NeoXX: IS THIS AI

Default avatar.png NeoXX: OR REAL

Default avatar.png FirstInLastOut: it's a discord

Default avatar.png FirstInLastOut: a real chat at least

Default avatar.png NeoXX: huh

Default avatar.png NeoXX: are my messages being sent to the discord?

struct: no

Default avatar.png NeoXX: oh

Default avatar.png svnnnnnny: hello

Default avatar.png NotSoClassy: hey babe

Tomi101: hey

jbyepez: hi

AlexPearson: heck yeah I actually finished one

jbyepez: :smile:

Default avatar.png soutenu: HEY