**Smelty dies
PatrickMcGinnisII: RIP
bheemu: ola
jacek: wooo chess as potw \o/
jacek: KelvinAndHubbles you need to advance to league to complete it
epistemologist_: i cant write fast code for my life
MasterCoderxD: http://chat.codingame.com/pastebin/eaf2422e-4e9c-41be-a7fb-4f05c258d0a9
ahthserhsluk: https://www.codingame.com/clashofcode/clash/17002497bb97d6d628f15b8cd3e3df6cfb50c36
MasterCoderxD: The above function still doesn't work for std::vector<std::string>
MasterCoderxD: Can someone make an edit?
MasterCoderxD: http://chat.codingame.com/pastebin/eaf2422e-4e9c-41be-a7fb-4f05c258d0a9
kedarnath: hello
DCoderOP: hlo
DCoderOP: go to clash channel and join clash
MasterCoderxD: http://chat.codingame.com/pastebin/eaf2422e-4e9c-41be-a7fb-4f05c258d0a9
N00b_er: Hello
MasterCoderxD: Help!!!
MasterCoderxD: Some C++ pro help me!
jrke: whats the help?
MasterCoderxD: See the above texts.
jrke: sorry i am not handy with pointers
MasterCoderxD: Oh ok.
MasterCoderxD: Np.
MasterCoderxD: Anyone else?
philipplenk: What exactly are you trying to do?
philipplenk: http://chat.codingame.com/pastebin/f0c2fa3f-a996-43d3-b96c-acc61abdd6d9
MXgms: hi
derjack: good morning
KiwiTae: MasterCoderxD looks like a mix with cpp and python lol
MasterCoderxD: Yeah that's what I was experimenting on lol.
derjack: cppython?
BlaiseEbuth: :nauseated_face:
derjack: (╯°□°)╯︵:upside_down:
BlaiseEbuth: :) ノ( º _ ºノ)
KiwiTae: BlaiseEbuth o/
BlaiseEbuth: \o KiwiTae
Thyl: Hi !
Uljahn: derjack: you mean cython?
derjack: dython!
Mr.pOoH: hey world yesterday i feel someone is controlling mouse then watched youtube and it says something problem with the touchpad :o
BlaiseEbuth: :thinking:
derjack: :mouse:
Mr.pOoH: :cat:
Swetsen: hi guys!
Mr.pOoH: :wave:
Hackercodehp: hi
sweety07: hii
kkkkl: hi
kkkkl: i just discovered this site
Ketan_3108: https://www.codingame.com/clashofcode/clash/170056852d42e9ce62e87022d97010809cef17f #clash
derjack: yhm
Alex-1: hii
AntiSquid: Ketan_3108 #clash
Alex-1: hii AntiSquid
Uljahn: he was not afraid to die, oh, brave Sir Robin
Alex-1: what is Spring Challenge 2021
struct: The next contest
derjack: *leaked* the theme will be ponies
Astrobytes: What's that? The them is 'Throw the brony down the well"? :P
Astrobytes: *theme
struct: I wonder what the game will be :thinking:
struct: Do you usually join contest derjack?
Prabahrn: hi
kkkkl: hi
Astrobytes: Salt mining? :P
derjack: struct i join to see whats up but at most i get silver
Kitkat260: goooooooooood morning coders!
derjack: did you see the puzzle of the week this week
Santjl: hello
Santjl: good morning my worldwide friends
Astrobytes: Chees
Astrobytes: Chess even
derjack: :cheese:
struct: Should I even paste the move generation on slcc feedback?
struct: I think the main reason my bot failed is because I didnt work on my eval
Astrobytes: Sure, if you think it adds something
struct: How can we pm in the forum? I wanted to pm tbali
struct: to ask him if he wants to join
Astrobytes: Find a post by him and click on his username is the fastest way I guess
struct: done
Astrobytes: Cool, nice one
struct: Im not sure the move gen adds much
struct: The only difference is that it takes into account 2 angles
struct: instead of 1
Astrobytes: Well, just give it a mention if appropriate then
Therabidpanther: help me stop being so tired
BlaiseEbuth: Go to sleep
Therabidpanther: work and school makes that impossible for the next 6 hours :(
BlaiseEbuth: Sleep where you are
Therabidpanther: i mean, this is possible... its just a cubicle lol
Therabidpanther: a cubicle at work where i do codingame.... lol
KiwiTae: I am your boss... coming to your cubicle right now
Therabidpanther: time to wake up i suppose
DonaldReza: hello everybody
struct: hi
sleepy-coder-101: hello world
sleepy-coder-101: i am sleepy
sleepy-coder-101: zzzzz.......zzzz......zzzzzz....
Kitkat260: hi
KiwiTae: o/
BlaiseEbuth: Oh. The break is over...
Kitkat260: ?
Kitkat260: i need help with coding
AntiSquid: we all do, ask Automaton2000
Automaton2000: but before i had a bad submit
Kitkat260: welll..... can u help
Mr.pOoH: :D
Mr.pOoH: Automaton2000 hi
Automaton2000: madknight i think you need to track the opponent
AntiSquid: https://www.education.com/magazine/article/fifth-grade-reading-comprehension/
struct: sry im also learning how to code
Marchete: Can someone help me with the descent?
sanath-cousin: h
struct: solve it on paper first
struct: or press hints on the left
Marchete: good idea struct
AntiSquid: for descent write a simulation, also collect replay data and analyse it first
BlaiseEbuth: Hey! I found a bug in onboarding, where can I report it?
Marchete: just make a new forum post and PM moderators
KiwiTae: stop making fun of newcomers laaa they gonna walk out :p
AntiSquid: who you calling new comer ? 1 vs 1 me KiwiTae, see who's brest
derjack: breast?
BlaiseEbuth: Brest?
bready-to-code111111: funy
AntiSquid: best *
KiwiTae: its the city i live in atm
Astrobytes: Actually Brest is appropriate
Astrobytes: see
Astrobytes: So Kiwi is officially brest
AntiSquid: typo, the letters are next to each other and i have chubby sausage fingers, stop making fun
Astrobytes: No really, he moved to Brest, bretagne, France
derjack: fix your -2.29 chess bot
BlaiseEbuth: Brest is in Taipei ? I should review my geography...
Astrobytes: Tell them KiwiTae
bready-to-code111111: France and Taipei are easily confused, the names are very similar.
BlaiseEbuth: True
AntiSquid: he moved to europe he said so
Marchete: probably if you compare it on Javascript
struct: lol
Marchete: True or False on JS is a leap of faith
AntiSquid: woah ... the JS haters at it again
Astrobytes: It's true though! (or is it? :D )
struct: Give me a reason to love javascript
AntiSquid: you can't force love
bready-to-code111111: scratch is best language
AntiSquid: and haters gonna hate
BlaiseEbuth: That's fun, I just left a Js hate session on Fr...
KiwiTae: I am currently working abroad in France
KiwiTae: just moved there this quarter
struct: how is it?
bready-to-code111111: is it warm?
BlaiseEbuth: Kouign Amann never hurted anybody.
Astrobytes: ^
AntiSquid: don't get the reference
KiwiTae: its cold as f
struct: Really?
BlaiseEbuth: Of course, he's in Bretagne, not in France
Astrobytes: It's always cold and wet in Bretagne
KiwiTae: its like 11°C
AntiSquid: it's colder everywhere this year
BlaiseEbuth: Virus fault
struct: This winter was hell in portugal
AntiSquid: had snow in april
struct: Houses are not made for this
BlaiseEbuth: Just build other houses
JordyH: Winter a hell in Portugal? It can't be that cold. It's not Finland
**BlaiseEbuth is sorry for this easy one
Astrobytes: It may be cold and wet KiwiTae but always remember en Bretagne, il ne pleut que sur les cons
BlaiseEbuth: :3
KiwiTae: :joy:
AntiSquid: how good is your french from 1 to 10 KiwiTae
KiwiTae: 9 I guess
derjack: omelette du fromage?
AntiSquid: how well do you understand hard-syllable-skipping french speakers?
KiwiTae: c'est la vie le paing ohlala
KiwiTae: AntiSquid no clue I haven't interact with people so far, lockdown , no social circle
AntiSquid: lockdowns are the reason it's colder probably, less carbon emissions, less green house effect
Therabidpanther: i suppose that could make sense
Marchete: so are you telling me I need to burn oil and stuff to avoid an ice age?
Marchete: makes sense
KiwiTae: its due to la Nina in the pacific, cools down the surface a bit
BlaiseEbuth: https://www.youtube.com/watch?v=QJwIsBoe3Lg
struct: the girl?
KiwiTae: still gonna make it top 10 of warmest years since 1700s
Astrobytes: yeah struct you have la nina and el nino
struct: "You reached the limit of plays for a period of time."
struct: How long is this cooldown?
AntiSquid: i am sure blaise wears a garlic bulb necklace daily
KiwiTae: bout 1y or so
Marchete: niña y niño
AntiSquid: nino
BlaiseEbuth: Of course
Astrobytes: thanks marchete :)
Marchete: https://link.springer.com/chapter/10.1007/978-3-319-09165-5_5 what a bargain!
struct: does astro even has that key?
struct: ~
Marchete: at that price I buy the top player from CG
Astrobytes: No, could have looked it up but I didn't
Marchete: and it will have much better ideas
Astrobytes: lol Marchete, quick someone tell MSmits
BlaiseEbuth: Who pay for ebooks
struct: Msmits about to get scammed again
Astrobytes: Smits pays for papers
Marchete: only 39.58
BlaiseEbuth: :scream:
Astrobytes: Well, he paid for one. We're just never letting him forget it :P
Marchete: CGermint never forgets
AntiSquid: if you like the cover! buy it!
then you can add it to your bookshelf, it helps you look smart
Marchete: "the code can beat humans!"
Marchete: did you read that!? huimans!
Marchete: humans*
struct: impossible
Marchete: like it was a f*king achievement with teraflops...
Astrobytes: what is this sorcery
struct: Just wait until you find that the secret to beat humans was to add a opening book
Astrobytes: hahaha
AntiSquid: you could invite the author to CG to advertise his book better
derjack: is this what reCurse bought to beat me at that game
BlaiseEbuth: reCurse bought nothing to beat you
derjack: except new PC
BlaiseEbuth: He bought a new pc to beat you ?
AntiSquid: are pc parts available again?
struct: no
struct: its even worse now
AntiSquid: GPUs nowhere to be found
BlaiseEbuth: Why do you need pc parts when you can have a complete one... :rolling_eyes:
struct: if I had a 3000 series I would sell it
AntiSquid: ya good profit struct
AntiSquid: because i am not paying extra for pre-built BlaiseEbuth
derjack: or preprint
BlaiseEbuth: You prefer don't pay at all and stay with your old pc.
AntiSquid: there are few companies that allow you to pick your parts and they build it for you, but even then i would be worried about them selectively leaving out parts that accidentally above average
lastshamp: :poop:
AntiSquid: it's like 6 months old geez, i don't know how i'll survive
BlaiseEbuth: :(
derjack: Marchete, bought it. fully recommend https://i.imgur.com/SHDIaQw.png
struct: o.o
Marchete: lucky you :D
Marchete: in fact I wasn't looking about breakthrough, but about MCTS-EPT
derjack: just replace the result of playout with eval, ezpz
Marchete: while(!winning()){ do_stuff_to_win(); }
Marchete: ezpz
derjack: for more, send $400 to 666-420-69-2137
inoryy: fyi: 0% of what you pay goes to the authors
struct: 0%?
Marchete: this can't be true
Marchete: no royalties at all?
inoryy: welcome to academia
Marchete: like almost 40eur for some paper and nothing for the authors?
BlaiseEbuth: Scientists are like artists, they're paid in visibility.
inoryy: it's quite literally 0.
Marchete: they are paid in prestige
Marchete: do you have a lot of prestige, inoryy?
Marchete: btw :)
inoryy: moreover sometimes you can get a free copy if you just email the authors, as they retain the rights
BlaiseEbuth: Or you can just use a downloader
inoryy: don't think so? how would I measure it : D
inoryy: I mean I'm not saying there's a simple way to bypass the paywall but I'm saying if you did use it then authors would literally lose nothing
Astrobytes: Yep. Researchers do the work so journals and sites can make money.
BlaiseEbuth: Pretty normal.
BlaiseEbuth: :money_mouth:
Marchete: I would never bother an author for a paper tbt
inoryy: https://twitter.com/hwitteman/status/1015049411276300289?lang=en
BlaiseEbuth: ez money
inoryy: also I'm not saying there's a simple way to bypass the paywall that you should use, but if there were then the book you linked would definitely be available through it
BlaiseEbuth: C'mon inoryy, everybody here knows the way. ^^
struct: Can confirm, I profited a total of 0€ from all the papers I published.
BlaiseEbuth: But how much prestige!
Marchete: prestige worldwide
struct: I think you are on to something
struct: you should buy that domain if its available
Marchete: https://www.youtube.com/watch?v=ciS914MaDl8
inoryy: Marchete not sure if you saw before, but I stumbled on your Mokka project -- nice work!
Marchete: I don't know what I was doing
Astrobytes: Well, if that's not high praise I don't know what is :)
Marchete: so I beg your pardon on stupid stuff I did in that repo
inoryy: pardon granted
BlaiseEbuth: Mokka is pretty good cake.
BlaiseEbuth: +a
Marchete: Convolution layers is a hell to do it in AVX2
Marchete: there seem to be some info around
Marchete: but damn
Marchete: are*
inoryy: yeah, can imagine
struct: I think re curse mentioned that avx512 is a little more friendly for NNs
Marchete: yes, I found a lot of info on higher AVX versions
Marchete: in most cases AVX2 is ignored
Marchete: even by Intel
Marchete: or it's overly complex
Marchete: so I gave up on that
MSmits: hey Marchete I am using your Mokka thingy to jumpstart my machine learning experimentation, so thanks a lot
MSmits: the fact that i know i can use it in a CG bot, means i can use TF
Marchete: I'm trying to use it, but still I'm working on the jacekmax
MSmits: ah ok, i've written jacekmax before
MSmits: it's very similar to mcts with ept
MSmits: except you backpropagate minimax style
Marchete: I'm trying to do a monster
MSmits: what monster?
Marchete: jacek+smitsi+max
MSmits: i am not sure that is possible
Marchete: hence the monster
MSmits: smitsi relies on statistics that are decoupled between agents. If you maximize, then you need exactness
MSmits: the whole idea about smitsimax is that the fuzzy statistics makes it work
MSmits: but if you maximize, you cant rely on that
Marchete: right now I don't trust the maximize part
MSmits: you can trust it if you have a single, exact game tree
Marchete: after N visits the code sees that another branch is much better
Marchete: and changes the whole max() part
MSmits: well I translated my mcts with ept to jacekmax in oware and got same rank
MSmits: as in above all non-nn, below the nn
Marchete: nice
Marchete: jacekmax without rollouts, right?
MSmits: mcts with ept seemed to be somewhat stronger, but eval was tuned for mcts with ept
MSmits: yes without rollouts, eval
Marchete: right now the jacekmax is about 30th, but my minimax is like 15th
MSmits: just eval children
Marchete: eval children is costly
MSmits: depends on the eval
Marchete: he takes out the policy part of the alphazero
Marchete: in exchange of more game simulations
MSmits: well its not really alpha zero
inoryy: > MSmits 04:30PM
> i am not sure that is possible
famous last words
MSmits: well in this case it just makes no sense to me inoryy, I didnt mean it is hard
Marchete: can't the negamax part suddenly find a trap and "invalidate" the best branch?
Marchete: like even if you have a lot of visits
MSmits: yeah, it can, but mcts can skip traps too
Marchete: I mean negamax inherently can fall in traps that change a lot the propagated score
MSmits: not really
Marchete: I thought it can happen the same on smitsi
MSmits: because it's min maxed
Therabidpanther: regex is the devil
BlaiseEbuth: No. I'm
Therabidpanther: oh
MSmits: so if it suddenly raises the score a lot, the parent will just not be picked
MSmits: so extremes will disappear
MSmits: the biggest problem is the horizon effect
MSmits: because the player who's turn it is might have an inherent advantage or disadvantage
MSmits: so if the dept between leafs is not the same
MSmits: some leafs may be p1 and some may be p2
MSmits: eval can correct this partially
MSmits: I usually have player-turn as part of my eval score
Marchete: do you prefer jacekmax or MCTS-EPT?
MSmits: but jacekmax, just like smitsimax has some features that makes you question why it would even work, but it does, I tested it for oware, jacek tested it everywhere
MSmits: mcts-ept
MSmits: because i know why it works generally
MSmits: jacekmax is a bit weird
Marchete: I don't like the negamax part
Marchete: I prefer sumScore/visits
MSmits: why not, mcts also has negamax
Marchete: the max() part
MSmits: oh ok
MSmits: well then it's just mcts with ept
MSmits: use that instead
MSmits: if you do sumscore/visit i mean
MSmits: it works fine if your eval is good
Marchete: fine?
MSmits: it's very simple to convert one way to the other, just a few lines of code
MSmits: similar i mean
MSmits: at least for oware
Marchete: fine doesn't mean good or excellent
MSmits: jacek says in some games, one was better and in other games, the other
Marchete: it's just okiish
MSmits: no, i mean i beat everyone with both algos
Marchete: top 30 or so
MSmits: except the NN's
MSmits: i can beat the NNs by cheating with a counter book though :P:
Marchete: do you save gamestates on MCTS nodes?
Marchete: or replay the whole thing
MSmits: in some games yeah
MSmits: actually in most
MSmits: if the state fits in 128 bit or less
Marchete: with an NN is harder to keep books
Marchete: no left
Marchete: no space left
MSmits: i doubt that, why not just make a smaller nn
Marchete: I was thinking that
Marchete: keeping in 128 bits
MSmits: which game is this?
MSmits: oware?>
Marchete: oware, yes
MSmits: my state is 64 bit :grin:
Marchete: oh :D
MSmits: well + score sry
MSmits: the board is 64 bit
MSmits: and then i have score also
MSmits: so another 16 bit
Marchete: I have tons of useless info on MCTS nodes
MSmits: my state is: 5 bit per pit = 60 bit
MSmits: then 4 bits left over
MSmits: for the max 1 pit that has > 30 seeds
struct: marchet you ever tried to calculate value of childs with avx?
struct: to get the best one
Marchete: hmm,no, not in boards
Marchete: maybe you can do it in sim games
MSmits: did you want to avx oware Marchete?
Marchete: noooo
Marchete: I wanted to do AVX for Convolution layers
MSmits: i think I have one of the fastest implementations for sim in oware, but no avx
Marchete: my sim is ugly and buffy
Marchete: buggy*
MSmits: I do a bunch of lookups, little ones (cache efficient)
inoryy: but does it slay vampires?
Marchete: I hate oware rules more than PCR sim
Marchete: oware = you can move, yes, but then it's invalid and game is ended
Marchete: you can sow but can't capture
Marchete: a lot of yes but no
MSmits: meh, i hate pcr more. But think about my boardstate and how i have 4 bits leftover for >31 seed pits, think about how complicated that sim becomes :P
Marchete: I prefer to use 128 bits for board + score
Marchete: 64+16 vs 128
Marchete: I don't know
MSmits: i like to have a cache of 40 million nodes :P
MSmits: and reuse tree, for the whole game
Marchete: how do you reuse nodes?
MSmits: i just check the state
MSmits: if it's the same as one of the children of the root, i take it as the new root
Marchete: you don't reuse nodes, then, no?
MSmits: yeah i do
MSmits: thats reusing nodes
Marchete: how do you keep track of that?
MSmits: the root is just the pointer to one of the nodes in my cache
Marchete: I splitted the big array of node on smaller parts
MSmits: when i play a move, i change the root
Marchete: like 1 per depth
MSmits: nah, why do that
MSmits: just make a nodes[40 million]
MSmits: or whatever fits
MSmits: let me share my node, sc
MSmits: sec
Marchete: I have like 6million nodes, without gamestate :)
MSmits: http://chat.codingame.com/pastebin/3522e412-71cc-4105-8b21-9524cedfa5c2
Marchete: Cache Manager. Size:347MB - 5700000 nodes splitted in 57 blocks
Marchete: you change turn, you search your move + enemy move
Marchete: have a match
Marchete: then you set that as root
Marchete: root->parent = nullptr;
MSmits: my nodes dont know their parents
MSmits: they dont need to
Marchete: how do you backpropagate?
MSmits: Node* selectedNodes[200];
MSmits: i store the pointer
MSmits: during selection
MSmits: http://chat.codingame.com/pastebin/183fb07b-3bdb-4893-8659-dca90372d5db
MSmits: i dont even backpropagate
MSmits: i forward propagate, if you mean the order in which i increment
MSmits: it doesnt really matter as long as you do the signs correctly
Marchete: it's a good idea
Marchete: I know
Marchete: but you don't "recycle" nodes,no?
MSmits: i think you overthink the tree reuse
MSmits: i do
MSmits: let me show you
MSmits: http://chat.codingame.com/pastebin/3d446e5e-f45d-4beb-b3ad-c11bd2a1ac09
MSmits: so i check all the children of the last root, see if any correspond with the current gamestate and then i set that node as the current root
Marchete: childMax is childCount right?
MSmits: yeah
Marchete: the current root I get it
MSmits: oh i also change the root when i play a move
Marchete: but you reorder or smth?
MSmits: so the root gets changed twice per turn
MSmits: no
MSmits: the nodes are just all over the cache
dudeOnGitHub: spam
MSmits: until its full
MSmits: then the whole thing gets reset
Marchete: what reset means
Marchete: lost all infor?
Marchete: info?
MSmits: yes, this happens when the full object pool is used
MSmits: you can also start at index 0 again and not reset
MSmits: and hope it doesnt yet need nodes that have been overwritten
MSmits: thats risky
Marchete: no
Marchete: that doesn't work
Marchete: trust me
MSmits: i used it before in uttt
RoboStac: thats how all my mcts bots work
MSmits: the reset happened at ply 20-30
Marchete: without reset?
MSmits: yeah he restarts and doesnt reset
MSmits: as long as none of the older nodes are still in use, its fine
MSmits: and you dont get to ply 30 on turn 1
MSmits: so thats usually ok
Marchete: I was splitting all nodes in different arrays of nodes
MSmits: the faster your bot get the riskier it becomes
Marchete: then mark the arrays unused
MSmits: nah dont do that
Marchete: it works but probably slower than yours
MSmits: for one, the gain of tree reuse is marginal
MSmits: it helps more in games with low branching
MSmits: in oware admittedly it helps more
MSmits: but still, the game has branched twice since your last turn
MSmits: think of how few visits will still be in that branch
Marchete: between 8% and 90%
Marchete: I got stats
MSmits: yeah in oware it is a lot more
Marchete: on my buggy bot
MSmits: but in uttt its pretty crappy
Marchete: Visits:3130/11746 = 26%
Alshock: un UTTT my reuse was about a few percents
Marchete: Visits:45738/52013 = 87% --> At that time I already lost :)
RoboStac: theres still enough turns in UTTT with only 1 or 2 good moves that it makes a big difference
MSmits: well there's also a price to pay for tree reuse, that you may not realize
MSmits: you get cache efficiency for not reusing
Marchete: why
Marchete: contiguous?
MSmits: because if you reuse, your nodes are much more spaced out
MSmits: yes
RoboStac: not massively, it's still going to spread out very quickly
Marchete: then yes
MSmits: i have no idea how strong this effect is
MSmits: but there must be some
Marchete: neither me, I tried the fastest reuse with low precached nodes
MSmits: in any case, having to reset halfway through the game is no big deal
MSmits: if it's just 1 time
Marchete: splitting in "levels" seemed the faster to me
MSmits: never tried this, but as i said, the smaller your nodes become, the less of an issue you have to solve
Marchete: but probably raw power and 1 reset can be better long temr
Marchete: term
MSmits: there are only a few games that require a reset
MSmits: bandas is notorious for this
MSmits: you do so many sims, you might run out
Kitkat260: which lang is easier to code?
RoboStac: bandas is a pain because it has infinite cycles and always has 4 moves
Kitkat260: ....
MSmits: yeah, but if you use an endgame book, most of those cycles disappear
Marchete: python3
Kitkat260: thx
RoboStac: yeah, but then I'd have to write an endgame book
MSmits: yes, i thought it was fun though :)
MSmits: got up to 3x4
Marchete: I'm trying to do some more generic MCTS, not ultra tuned for certain game
MSmits: darkhorse always uses the same generic mcts
Marchete: I see the no parent interesting
MSmits: i always end up with a different mcts, because i somehow go with the input structure
JBM: thor keeps going off the screen
struct: generic code is good if you want something coded fast
JBM: this is too hard
MSmits: sometimes you're always p1, sometimes the index is given etc
Marchete: print -3*vel, it works fine for thor
Marchete: you can do overloads
Marchete: I'm trying to do that
ZERIH08: hi
Marchete: but for the most performance
Marchete: it's not the best
MSmits: btw Marchete, oware also has endgame books
MSmits: I go up to 9 seeds comfortably, 10 seeds with some crazy workarounds
MSmits: and they are not stored in codesize
MSmits: so goes easily with nn :)
Marchete: "and they are not stored in codesize"
MSmits: nope
Marchete: what that means?
MSmits: 500 ms
MSmits: for a 9 seed db
MSmits: fully calculated
Marchete: ahhhhh
MSmits: also including the game turn as a variable
MSmits: so we're talking like 10 million entries
MSmits: it may give a different result on turn 190 than turn 180
Marchete: but then you don't have space for MCTS nodes
MSmits: why not, I store 1 result in 5 bit
MSmits: -15 to +15 seeds
MSmits: so dont need more
Marchete: tbh I don't use endgame books at all
Marchete: and barely any opening book
MSmits: opening books arent that great really, unless you specifically counter a deterministic opponent and thats cheesy
MSmits: endgame books in bandas are strong and also help in oware
MSmits: i am doing a uttt opening book experiment now, doing cg bench test every 1 million meta mcts games
MSmits: so about once a week
MSmits: so far, 3 million games, no net gain
MSmits: except for 1 guy, who it turns out, also has an opening book :P
Marchete: I prefer to copy your openings :D
Marchete: just clone all your moves and good to go
MSmits: this works
Marchete: that's cheesier :D
MSmits: when i still had my othello counter book that always drew games against oldjohn, you could just copy 1 game of us and play the same moves vs me, guaranteed draw
Marchete: :thinking:
MSmits: basically perfect play from both sides
MSmits: but just for that specific line of play (and abount 1k other lines)
MSmits: about
MSmits: but as i said, i turned it off, now i am just generating the book by uct, to get rid of the guilt of this cheesy way of playing :)
struct: All the games I ported are bookable :(
struct: I need to move amazons from wip
struct: I think
Marchete: how do you tune C and FPU on MCTS, Smits?
Marchete: I struggle with these
MSmits: FPU?
Marchete: first play urgency
Marchete: visits == 0
MSmits: C is just 1.0 usually, very rarely it's 1.2 or so
Marchete: or do you just play any visit==0 child?
MSmits: i havent gotten any use from fpu, i dont think i did more than 1 experiment though
MSmits: yeah i tried random first
MSmits: or first first
MSmits: works the same unfortunately
MSmits: also, if you eval all children on every expansion, then there is never a visits = 0 child
MSmits: they all start at 1
Jeanpier: gaaa
MSmits: i eval all on oware, bandas, onitama, yavalath. Not on uttt though
Marchete: I put all those in visits ==0
MSmits: when i eval all children the children also get their own score
MSmits: thats technically a visit
MSmits: so it's visits = 1
Marchete: hmm, but it's not expanded
MSmits: no, so i expand at visits = 1
Marchete: ah, visits==1 expansion
MSmits: yeah
MSmits: btw you can take this further
MSmits: darkhorse in bandas, expands 16 children
MSmits: so he skips a layer
MSmits: because you have 4x4
MSmits: i tried this, no benefit for me
MSmits: also very messy
Marchete: like DUCT and similar
MSmits: yeah thats also messy
TRAFFICCONE: hello
MSmits: hi
TRAFFICCONE: how are you
MSmits: I'm good, you?
TRAFFICCONE: good
MSmits: great
struct: Astrobytes and Scarfield any of you here?
Marchete: Assertion `child->parent == this' failed.
Marchete: great!
Scarfield: yea
MSmits: huh, why would something be its own parent
struct: Tbali said he can join wednesday or thursday for escape
MSmits: isn't there a time travel movie about this
struct: How would it work MSmits?
Scarfield: kewl, i should be able either of the days
MSmits: struct, i dont mind explaining, but i dont want to ruin your innocence
BlaiseEbuth: "Predestination" MSmits
MSmits: it;s a really f-ed up movie
MSmits: yeah thats it
MSmits: google it if you want to struct
struct: oh its for 16+, I cant watch it
MSmits: better if you don't :)
Scarfield: Futurama Fry is his own grandfather iirc
DestinyBrown: hey
BlaiseEbuth: The guy is His own father and mother
MSmits: yeah
struct: spoilers
MSmits: well
MSmits: i guess so
MSmits: it's a pretty old movie though
MSmits: well 10 yrs maybe
AntiSquid: there's dota cartoons on netflix lol
BlaiseEbuth: Yeah. 8 years
MSmits: i saw that AntiSquid
MSmits: a few episodes
MSmits: they were ok
MSmits: if you're bored...
a.fitzy98: Hey, where do you find details on the challenge?
MSmits: it's secret
MSmits: so you can't prepare
BlaiseEbuth: shhh
MSmits: you just know it's 10 days and its a bot game
a.fitzy98: Ohh I thought it was already released
a.fitzy98: ok cool thanks
MSmits: oh 11 days now
inoryy: I mean you can still prepare by doing multis if you've never participated before
MSmits: yeah, but dont pick the really simple multi games, contests are never that simple
struct: You should prepare :zipper_mouth:
MSmits: preferably pick the ones that need some pathfinding, like bfs
Astrobytes: struct and Scarfield: Thursday sounds good for me
struct: Same time as last time?
BlaiseEbuth: Ze time
MSmits: hey , what are you guys doing
Scarfield: sure, how about asking zeno, he seemed interested?
BlaiseEbuth: Pew pew!
a.fitzy98: By multi do you guys mean multiplayer games? And the task is to create a bot to play the game? I have never taken part in anything like this before
struct: Sure, the new escape is 2-6 people
MSmits: a.fitzy98 yes
struct: MSmits Im doing bot programming
Uxie: I have no idea how to code or read code... where do I start? I'm looking at the easy puzzles but I have no idea what I'm reading
a.fitzy98: bf = battlefield? Thanks btw
BlaiseEbuth: ^^
inoryy: :joy:
Scarfield: bfs breadth first search ;)
MSmits: a.fitzy98 you could take a look at: https://www.codingame.com/multiplayer/bot-programming/fantastic-bits
BlaiseEbuth: battlefield is cute
a.fitzy98: I am going to do well in this I'd say :joy:
MSmits: or https://www.codingame.com/multiplayer/bot-programming/spring-challenge-2020
Astrobytes: what time was last time struct? (I already forgot lol)
BlaiseEbuth: Ze time
struct: I think it was 8 CET
a.fitzy98: Thanks guys lol
inoryy: don't worry, literally everyone you're talking to started from your position at some point, a.fitzy98
Scarfield: ^
Astrobytes: ^this
struct: or 20 GMT
MSmits: a.fitzy98 if you have some programming skill, just the basics, you can usually pass the first wood leagues and have some fun doing it
Astrobytes: Yes that's cool for me struct
Scarfield: me as well
BlaiseEbuth: Of course
a.fitzy98: Cool will give it a go cheers
struct: ok I will ask tbali
Astrobytes: cool beans. I can do pretty much anytime that day if he can't make that time
ZarthaxX: howdy
MSmits: hey
ZarthaxX: hey smito :*
struct: hi ZarthaxX
ZarthaxX: many ppl today
KiwiTae: ZarthaxX and meeee
MSmits: zarthy
KiwiTae: o/
ZarthaxX: hellou guys
KiwiTae: MSmits hellou
Scarfield: Zarthoo
MSmits: hi KiwiTae
ZarthaxX: structo :* kiwo :*
ZarthaxX: scarfoooooooooooo
KiwiTae: that bear is famous
ZarthaxX: im a fox
BlaiseEbuth: What bear
BlaiseEbuth: Zarthouille!!!!!
MSmits: you were a bear before werent you?
KiwiTae: #bringbackthebear
ZarthaxX: BlaiseEeeeeeeeeee
ZarthaxX: i will bring it back for contest :P
ZarthaxX: yes smito :rofl:
MSmits: as if i needed yet another reason to be in this contest
MSmits: cant miss the bear
BlaiseEbuth: Yeah, the t-shirt lack was enough
Astrobytes: ZarthoxXy o/
ZarthaxX: :rofl:
ZarthaxX: hey AstroBit
Scarfield: AstroNibble
ZarthaxX: :rofl:
Astrobytes: BistroBytes
MSmits: Astrobytes, do we need to explain the game to you
MSmits: you're supposed to give *them* names
Astrobytes: Yeah I know, I like to compete
MSmits: mmh ok :)
Scarfield: trying to 1up a name is acceptable
Astrobytes: ^ see, Scarfupmanship here gets it
Scarfield: though the rules admittedly is unwritten
Scarfield: are*
Astrobytes: Who needs rules. Or statements :P
Scarfield: not ceg
MSmits: he does
H7: What kind of Spring Challenge 2021 contest?
MSmits: he has a rulearray
Astrobytes: :rofl:
MSmits: H7 unknown, just that it's probably a bot game
Astrobytes: H7: Bot programming is all we know for now
MSmits: you'll get maybe 1 or 2 hints from a trailer, the days leading up the 6 may
MSmits: up to
Smelty: o . o
struct: Can it be a optimization one?
MSmits: no idea
MSmits: usually it's bot game, i dont know if they would tell us if it was optim
MSmits: the last optim was a weekend contest
inoryy: you can treat bot contests as optim, don't let society limit you
Astrobytes: :grin:
MSmits: yeah, the opponent is just another variable
MSmits: just optimize them away
inoryy: this is the way
Astrobytes: The real Zen of botting.
MSmits: hey inoryy, are you involved at all with muzero?
MSmits: i thought it is quite cool
inoryy: MSmits nope, not involved at all
MSmits: ah ok. I really like how a paper gets shared with some pseudo code and then implementations pop up all over that you can just try out for yourself
MSmits: i get turned off by all those package requireements and such, but the fact that it's accessible at all is nice
inoryy: yeah, it's really cool to see community excited enough to re-implement it :)
MSmits: really shows you the power of an idea. The fact that something is possible is almost enough to recreate it
MSmits: tell someone x can be done and they will do it
jacek: but not enough power to do it :(
MSmits: well thats true, but you can do it on a smaller scale
MSmits: a small environment requireing a simple network to understand
MSmits: muzero is so special, because a muzero bot will actually sim the game in a way that's not consistent with the game rules and this may lead to better result
MSmits: just like when you throw a ball, you dont know the exact physics calculations, but you model of physics is good enough
MSmits: your
Scarfield: assuming i can hit something with a ball? :p
MSmits: yeah, sorry, i meant in general. There are exceptions
Scarfield: xD
Scarfield: but it has to at least indirectly learn the rules, to play by them? i could easily win at chess if i just ignored the rules i mean
MSmits: I think it knows the possible moves to play on the current turn
MSmits: it gets told that
MSmits: but it doesnt know what happens after thart
MSmits: the model is learned concurrently with the policy and value parts of the network
MSmits: so when it builds the tree, it will often not be entirely correct
MSmits: some moves will be so bad, it might consider them not part of the possible set of moves
MSmits: (or something...)
Marchete: do you think muzero has any use on CG?
Marchete: I think it's just for bigger environments
MSmits: might be too complex for cg, but we've been wrong before
MSmits: there's free implementations of muzero
MSmits: you can download them and test for connect 4
Marchete: I see alphazero
MSmits: and also othello i think
Marchete: for CG
Marchete: but not muzero
MSmits: i found some, let me refind sec
Marchete: but what NN size?
MSmits: https://medium.com/applied-data-science/how-to-build-your-own-muzero-in-python-f77d5718061a
Marchete: sometimes they have connect4, the 6x7 or whatever the standard size
inoryy: wrt compute / power requirements you might find this paper interesting https://arxiv.org/abs/2104.03113
Marchete: and 5MB NN
inoryy: scaling laws is pretty big topic in deep learning in general atm
MSmits: thats cool inoryy
jacek: could simple MLP be used instead of convnets in muzero?
Shree007: I am new to coding game
Shree007: How to play it
Marchete: free paper! :D
MSmits: dont know enough to answer that. But convnets are good because parts of your board interact with neighbours more strongly than with far away parts of the board, right?
MSmits: so you use filters similar to what is done with images?
MSmits: maybe i got this wrong
**Marchete looks at inoryy_
jacek: so far MLP is good for my purposes
MSmits: well MLP should be able to do anything, but it might not do some tasks as efficiently as convnets
inoryy: muzero isn't tied to NN architecture, though you'd probably struggle to reproduce image-based results like the Atari from pixels performance
MSmits: so it might work well with oware
jacek: just like alphazero i suppose then
MSmits: convnet does not make sense to me with oware
jacek: im using a0 like without policy anyway
Marchete: indeed
jacek: so maybe next step will be mujacek
Marchete: I don't really see the benefit on CG
MSmits: that's a jacek with no human input and no idea about the game rules
struct: By the time I learn NN the new thing will be released
Marchete: maybe put that muzero on a challenge, go to sleep and in 5 day you have a winning bot
MSmits: Marchete from what i understand from the papers, muzero has similar performance to azero on games where azero has been tested
Marchete: days*
MSmits: the strength is in other applications
MSmits: like, building a model of an unknown environment
MSmits: like a robot exploring a space
Marchete: that's what I'm trying to say
MSmits: yeah, you're right
Marchete: CG has strict, written rules
Marchete: no open spaces and that
inoryy: muzero shines as a model-based agent, e.g. where env simulation is very expensive or even impossible
Marchete: maybe as a proof of concept
Marchete: who knows
jacek: well its about seeing if it works
Marchete: I see muzero working on PCR
Marchete: tons of chips
MSmits: yeah might be nice
Marchete: where "image" is important
MSmits: or fantastic bits
jacek: you can do perfect tic tac toe with few ifs, but if the goal is to learn minimax, you'd use minimax there
Marchete: I learn what I can make use of it
jacek: greedy
Marchete: I think I can use alphazero, actor critic or similar
Marchete: for muzero I see little use
MSmits: in board games, yes
Marchete: in CG restrictions
MSmits: might be good for chaotic, hard to sim games maybe?
MSmits: maybe meanmax ?
Marchete: I think a0 does it
MSmits: yeah i guess so
ShortBaited: Automaton2000 say something funny?
Marchete: maybe better with Conv layers
Automaton2000: but u can get to a higher league
MSmits: I still have a nonworking smitismax for meanmax i never finished
MSmits: the algo should work
struct: Other player actions dont affect your possible actions?
Smelty: o.O
MSmits: they affect the result of your actions
MSmits: but not your possible actions
MSmits: like csb
struct: I see
MSmits: you can get pushed around but still get to use the same actions
struct: Yeah, I thought it was like FB
MSmits: its not 100% sure though
MSmits: there are some effects from special abilities
MSmits: so you make some assumptions
Marchete: the hard part are the "spells"
MSmits: yeah i would just allow them at depth 0
Marchete: yeah
MSmits: and not in deeper layers
Marchete: like me
MSmits: i do the same in csb
struct: you do smitsi on mean max marchet?
Marchete: I always have a different children creation
Marchete: for turn ==0
Marchete: no
Marchete: only on Xmas and PCR
MSmits: me in pcr, xmas, csb and an early version of it in kutulu
MSmits: should improve that version one day
MSmits: it's a C# bot with a bad eval, still got 4th :)
Marchete: kutulu was the first one, right?
MSmits: yeah
MSmits: i liked your post about the yoga mats btw
MSmits: i got the same search result
Marchete: based on a true history
MSmits: i was gonna say something about google thinking you're interested in yoga
MSmits: then i did the same search and it showed me that too
jacek: :?
MSmits: if you google uct forest jacek
MSmits: daporan used it in his contest win
MSmits: it's the same as smitsimax basically, but he calls it that, which is fine. It's more descriptive
jacek: University of Cape Town?
MSmits: just multiple uct trees
Marchete: that's you, jacek?
Astrobytes: Yeah, I kinda like the name. Descriptive as you say MSmits
jacek: uhm
Smelty: */kill Smelty
**Smelty dies
Astrobytes: But you could call it Johnny's Mega Thing or something, wouldn't matter much
MSmits: they have hotlines for that Smelty
MSmits: well it would matter a little bit. It would tell me johnny is insecure about the size of his thing
Astrobytes: Trust you to lower the tone :P
MSmits: :P
struct: "Premature optimization is the root of all evil." -- Donald Knuth :(
struct: Even Stackoverflow is making jokes to me
MSmits: yeah, you're probably the only one who does this even worse than i do struct :P
Astrobytes: And you live by it struct!
struct: I was googling something and that shows in the answer ...
Astrobytes: It's a sign
MSmits: yes, google is getting better
MSmits: it showed us yoga mats earlier
Astrobytes: Messages from Knuth.
MSmits: knuth will always be the dancing links guy to me
MSmits: what a great algo
jacek: or dancing sort
MSmits: dont
MSmits: dont link it
Andrew-H:
Andrew-H:
Andrew-H:
Andrew-H:
Andrew-H:
Astrobytes: Andrew-H: stop it or get kicked
jacek: :(
MSmits: oh fine, link it
MSmits: but i am not clicking it!!
jacek: https://www.youtube.com/watch?v=dQw4w9WgXcQ
MSmits: damnit
MSmits: i thought you were going to link the dancing folk sort thingy
MSmits: my colleague always makes links like that in our teaching materials
MSmits: it will say: "The answers to the next test: here"
jacek: id like to meet him
MSmits: he's fun
Astrobytes: Not very antisocial of you jacek
Busting: wasup g
MSmits: I can link a YT video of him explaining how to code a python TTT bot
jacek: :angry:
MSmits: i mean TTT game btw, not bot
Busting: what yall talking about
Astrobytes: That's better jacek
MSmits: various things
Busting: such as?/
MSmits: yoga mats
Astrobytes: Stuff. Things. Suchlike.
Busting: sheeesh
Busting: i had a yoga matt once but my fish ate it
MSmits: dont store yoga mat in fish tanks
MSmits: also i did not mean a yoga named matt
Busting: i didn't knew
Busting: i call my yoga mat matt
Andrew-H: Do you guys know Candice?
Busting: fuk u
Busting: no andrew
Astrobytes: Ban next time.
jacek: what a game https://www.codingame.com/share-replay/541648486
MSmits: nice one, I noticed games between good players last a lot longer
jacek: its somewhat bugged
Busting: ok im more civil
Busting: jacek wanna know my favorite tea?/
jacek: earl grey, hot?
Busting: you shawtea\
Astrobytes: Jeez.
MSmits: I didn't get it, but I guess thats good
Busting: ty
Busting: do u get payed good if u have a job to code
Astrobytes: I've never coded a job.
MSmits: better than most jobs, worse than some
MSmits: produced a lot of jobbies though
Astrobytes: :rofl:
Astrobytes: Well played sir, well played
MSmits: 1 jobbie a day keeps constipation away
Scarfield: xD
Busting: lol
Astrobytes: hahaha
Busting: hes not wrong
Kitkat260: hi
Busting: sup daddy
Kitkat260: how r uu
Busting: gybtfr5v486 79o
Kitkat260: ggjja78jgy
struct: stop spamming
Kitkat260: ???
Busting: its not spamming if i did it once
Astrobytes: Busting is already on a ban warning
struct: you were talking gibberish
jacek: Automaton2000 ?
Automaton2000: you dont even know what that means
Busting: astro y expose me like that
Kitkat260: actally
Busting: imaging being mod
helenabjalkovska: keysmashing isn't spam
Kitkat260: Lmao
helenabjalkovska: mod abuse
Busting: imagine*
MSmits: mmh you're going to unleash the astro soon =/
struct: oh no mod shaming
struct: What will I do
Astrobytes: ikr
Busting: can mod mute
MSmits: mod can ban
MSmits: which is the same
struct: yes I can mute you
MSmits: more or less
struct: ^
Busting: lol weak power
struct: not as weak as your bait
Astrobytes: ^
Busting: thats just sad
Busting: ur a mod how dare u
Scarfield: he already has a warning, why postpone it?
Busting: weres ur boss i need to talk to em]
Astrobytes: Go right ahead.
kresteodymium: hello world
Busting: who yall boss
struct: Ill give you my boss email
kresteodymium: hi Busting
struct: You should see it soon
kresteodymium: hi struct
kresteodymium: hi Astrobyted
kresteodymium: *s
kresteodymium: wassup
MSmits: kresteodymium, type ast then hit tab
kresteodymium: o nice
kresteodymium: didnt kno that
kresteodymium: thx!
MSmits: np
kresteodymium: :)
struct: Some soccer drama
struct: I guess I know what the news will be on the next few days on portuguese channels
Astrobytes: European Super League stuff?
struct: yes
Astrobytes: Saw that earlier.
Glad I don't bother following the game any more.
jacek: soccer eh
struct: TLDR: top teams created a new league to milk money
Astrobytes: Yep. That's it in a nutshell.
struct: No German or French teams though
struct: Not sure if any will join or not
kresteodymium: soccer is epic
kresteodymium: *football
kresteodymium: lol
Myco: Hello?
ShannonNorris97: Has anyone got any decent resources about iterative deepening with minimax? The stuff that I'm finding seems completely different
jacek: hm?
jacek: what did you find
ShannonNorris97: I'm just skimming to find some decent resources, but it just seems like it's own algorithm completely, so I'm not sure if I'm looking at the right thing
jacek: well its more general for DFS search
ShannonNorris97: Ah okay, and then it combines with the minimax algorithm in some way?
jacek: yes, as minimax is some sort of DFS anyway
jacek: I dont know about decent resource for that
Astrobytes: doesn't Carolus have an ID section?
jacek: maybe page 16 https://homepages.cwi.nl/~paulk/theses/Carolus.pdf
Astrobytes: :)
ShannonNorris97: Thank you - I'll take a look :)
jacek: sounds trappy https://athena.ecs.csus.edu/~gordonvs/papers/trappy.pdf
Astrobytes: Oh, don't think I've read that paper jacek
jacek: neither did i, just found some random minimax things
jacek: yahoo, that gives some memories
Astrobytes: lol indeed
busting2: how do i get unban
jacek: delete system32
SmallRad: how do i delete system32?
KiwiTae: SmallRad
SmallRad: ye?
KiwiTae: open shell and : rd /s /q "path"
KiwiTae: but i dont recommand
SmallRad: oh ok
SmallRad: yolo
struct: o.o
KiwiTae: :innocent:
busting2: struct
busting2: unban busting
Astrobytes: /ban
busting2: no
busting2: dont ban me
busting2: ban yourself
struct: I tried
busting2: :rolling_eyes:
struct: You asked for the email I got you the email
BlaiseEbuth: Ah yeah. You're banning people without invite me !
Zenoscave: How's the ban hammer working
BlaiseEbuth: Pew pew!
Zenoscave: pewpew
Smelty: pewpew \===== - - - - - - - - -- - -
Secret_sister2: ( う-´)づ︻╦̵̵̿╤── \(˚☐˚”)/
**Smelty was killed by Secret_sister2 using an assault rifle
dannybai2020: pew pew pew hahaha
**Smelty respawns
Smelty: (⌐▀͡ ̯ʖ▀)︻̷┻̿═━一-
Astrobytes: wtf
Astrobytes: pewpewpew Zenoscave
Zenoscave: pewpewpew Astrobytes
Astrobytes: BlaiseEbuth: sorry dude, ping ya next time
KiwiTae: pezpez zeno
Zenoscave: zepzep KiwiTae
KiwiTae: failed
KiwiTae: xd
KiwiTae: you guys solved this one? https://www.codingame.com/ide/puzzle/divisibility-of-fibonacci-numbers-sum
Zenoscave: not yet
Zenoscave: 1st gold C4l
ZarthaxX: nice pewpew
Scarfield: https://youtu.be/fl0wIdGxfbQ?t=3
Zenoscave: thanks Zarthie
KiwiTae: noice
Astrobytes: oh cool zeno, what are you doing? Minimaxing approach?
Zenoscave: ok bye all I gotta work. Nope FSM
Astrobytes: Ah right, thought there was something special going on :D
Astrobytes: Laters man
Zenoscave: Not yet :) Might sim later in Legend
Astrobytes: Seems doable. Fiddly game it is.
ZarthaxX: scarfo LOL
Scarfield: xD
Astrobytes: StarWarsPewField
Astrobytes: PewthaxX
Scarfield: IncomeTax
Scarfield: copyPasteroBytes
Astrobytes: What did I copypasta?
Astrobytes: "IncomeTax" :rofl:
Scarfield: nothing(?) i just got the idea for the name
Astrobytes: It makes my teeth itch. PastaBytes I can live with
Astrobytes: Or PastryBites
Scarfield: PastryBites xD
Scarfield: i wish you could give others nick names, so you would see that instead of the actual name in chat, you would be PastryBites
Astrobytes: :rofl:
Sparky610: Hi
jacek: good evening
AntiSquid: itchy teeth? honey comb candy ?
Astrobytes: https://letmegooglethat.com/?q=makes+my+teeth+itch+meaning
AntiSquid: well it literally made my teeth itch whenever i tried it lol
jacek: those british teeth...
Secret_sister2: (ノ-_-)ノ ~┻━┻
AntiSquid: european
Zenoscave: Damn European teeth
Zenoscave: WFH is hard
Astrobytes: lol, distracted much zeno
Astrobytes: AntiSquid: 30 second review from you please - Neon Genesis Evangelion
jacek: WFH?
AntiSquid: my first day of WFH after a while lol
Astrobytes: working from home
AntiSquid: working from home
jacek: its about puberty ~
**AntiSquid tumbleweeds and crickets
Astrobytes: That good eh
Astrobytes: Thanks. Someone recommended it.
Zenoscave: https://images.app.goo.gl/kNtuzw6bT2rA9FrAA
AntiSquid: ah was replying to jacek's puberty joke, Astrobytes
Zenoscave: just a little distracted
Sparky610: aaaaaaaaaaaaaaa
jacek: joke?
AntiSquid: as for the anime, it's some emo kid who gets bullied and for some reason has social anxiety even if not bullied, the ending kills all the story for me, they fight hard for nothing @_@ Astrobytes
Zenoscave: AntiSquid Cruel Angel's Thesis is playing in your head now
Zenoscave: yw
AntiSquid: ah you replied to neon genesis evangelion jacek ? sorry, thought you were refering to the WFH thing
Astrobytes: Sounds pointless. On my don't bother list
Zenoscave: Serial Experiments Lain is much better
Zenoscave: ^ Astrobytes
AntiSquid: i agree with what jack said btw, it's true, Astrobytes
AntiSquid: i don't get the point of serial experiemnts lain ... imo better watch Stein's Gate;
Astrobytes: Yeah, figures. Thanks for the opinions guys, knew I could trust you all
jacek: try boku no pico!
Smelty: . 0 .
AntiSquid: lol
AntiSquid: change avatar orientation so he doesn't recognize you after watching it, jacek
Astrobytes: eh naw, you're alright thanks
Zenoscave: jacek wtf are you mentioning piko here for
AntiSquid: if Automaton2000 made an anime it would probably have story elements from boku no piko
Zenoscave: we have *standards* here
Automaton2000: how many languages do you know when i will be
YaoKim: yo dudes if i wanna sort descending, should i use > or <
jacek: isnt this anime to introduce to people who have never watched anime
Zenoscave: depends on what is on which side of the >/< YaoKim
AntiSquid: i think the moe stuff is bad enough already
jacek: what language?
AntiSquid: what did you watch so far Astrobytes that you liked?
AntiSquid: and neon genesis evangelion has amazing opening for sure
AntiSquid: the theme song is all it has
jacek: opening? ms would love it
jacek: also, maybe youll find something here https://myanimelist.net/animelist/ashiren
AntiSquid: is that your MAL account?
Marchete: lol
jacek: eeyup
AntiSquid: where's your overthetop CSS styling ?
AntiSquid: they removed mine
jacek: the what
Astrobytes: Nothing really other than random mecha stuff, Akira, and er the Project A-Ko series thanks to an old friend back in the day (a programmer)
I'm not into it really. But good is good.
AntiSquid: you didn't see many MAL lists did you? everyone has some outrageous flashy style added to their lists
jacek: meh
BlaiseEbuth: Mine was not flashy
AntiSquid: this one is mild for example : https://myanimelist.net/animelist/Cynder360
AntiSquid: (not mine)
AntiSquid: i gave lots of 1 ratings, kids would rage at me here too if i share mine
BlaiseEbuth: You did'nt like nothing
jacek: i dont rate, i just save for sharing
AntiSquid: Astrobytes watch Shokugeki no Souma and Yakitate Japan
Zenoscave: Has anyone suggested Gurren Lagan yet.
Zenoscave: Just to watch people shift uncomfortably
BlaiseEbuth: Kill la kill!
Astrobytes: Those look awful AntiSquid
BlaiseEbuth: Food anime... Why do I feel this is a little targeted ? :thinking:
AntiSquid: i thought so too until i've watched them Astrobytes
Astrobytes: Also I was exposed to Legend of the Overfiend
AntiSquid: haha
Zenoscave: I'd rather play code ala mode
struct: hard choice
BlaiseEbuth: tentacles...
struct: ah
struct: You could have said that first
AntiSquid: Mob Psycho 100 has awful graphics, but great story, so ..
Astrobytes: Urotsukidōji vs CalM? CalM, even if I have to play it for the rest of my life.
AntiSquid: Terra Formars has great graphics but one of the dumbest plots :D
AntiSquid: i rated 10 that one though :thinking:
Astrobytes: Anyway, enough of this weeb stuff.
Astrobytes: :P
AntiSquid: woah !!! let's come up with some slurs for western cartoons too, for equality
Astrobytes: whitey toons
AntiSquid: Aeon Flux was great btw
AntiSquid: ugly but good story
Astrobytes: Yeah I think I know that, even if I didn't watch it all
AntiSquid: Afro Samurai is about a black guy, created by jap studio and has western style to it
AntiSquid: 10/10
Astrobytes: sounds odd
JBM: 10/10=1
JBM: yup, it's odd
AntiSquid: lots of gore, great story imo
Scarfield: makes me think of Beverly Hills Ninja for some reason
Astrobytes: oh hey JBM, didn't see you for a bit. How's it going?
JBM: same old
JBM: maybe more bored
AntiSquid: your avatar is also a 10/10 JBM
Astrobytes: Yeah, I feel that
JBM: been streaming the shit out of the inlaw's ISP
Astrobytes: Hah, visiting time is it
JBM: how was sf42?
Astrobytes: was actually quite fun
Astrobytes: Wish I'd joined earlier
Astrobytes: Scarfield: respond to your forum tag
struct: It was good
struct: Just no time
AntiSquid: trying to better prepare for contest: staring at tryangle game and switching to different tab to do something else, been doing this for a while now, it's the graphics @_@
struct: Legend or delete?
AntiSquid: there is no legend
Astrobytes: Only memories.
Astrobytes: struct: When do we get Amazons boss
struct: no time atm
Astrobytes: Should I make one?
struct: Maybe after the contest
struct: If I dont have to delete
struct: if you want to
Astrobytes: What kind of boss do you want? Low depth alphabeta?
struct: Whatever beats random
Astrobytes: Cool. Will come up with something
struct: Thanks
Astrobytes: No worries
Astrobytes: Scarfield: Thankfield
Scarfield: np, didnt know of that feature, hardly ever use the forum, but nice
Astrobytes: Yeah, it's not the best forum system tbh, some things are less than obvious
Astrobytes: It works though I guess
struct: omg
struct: Im stuck in a loop
Astrobytes: A thought loop? You been at the LSD again?
Astrobytes: (kidding ofc)
struct: Cant answer
Zenoscave: When do you think a w/r is enough to ensure you'll beat the boss???
Astrobytes: Aha. Gotcha.
struct: 100%
Zenoscave: 55%? 60%?
Astrobytes: 50+ against the boss I thiiiink?
AntiSquid: depends on the bots that are below the boss too
Zenoscave: true Anti
Astrobytes: Yeah, you need that certainty above anything else
Astrobytes: Ohh on the LSD note, happy Bicycle Day
MSmits: Zenoscave, anything that is above 50% is enough. Mathematically, you will simply promote anyone else that you have <50% winrate against
Zenoscave: Makes sense
MSmits: but you have to actually be certain it is above 50%. 10 games in a submit means nothing
MSmits: do a 1000 game CG bench and have it be above 53% or so
MSmits: then you can be fairly certain
Zenoscave: So N > 100 at least?
MSmits: depends on the margin you have
Zenoscave: gotcha
MSmits: if in 100 games you have 60% WR thats pretty good also
Zenoscave: Binomial P values and what not
MSmits: yeah, i go by gut feeling here, done enough cg bench to know to be careful
MSmits: theres free confidence calculators though
Astrobytes: stahp
Zenoscave: I have a physical one too
Zenoscave: ti-34X
Astrobytes: It's like being in biology lectures again
MSmits: ti-30Xa
Zenoscave: maybe that's the one I have
MSmits: 7Euro 99
MSmits: when i got it
Zenoscave: 35 USD :/
MSmits: ripped off
MSmits: but you might have something different
MSmits: mine is not graphical
MSmits: just a simple scientific calculator
MSmits: it has sine and stuff
AntiSquid: maybe it's import tax for the US
MSmits: blame trump
Zenoscave: Mine has basic stats stuff
Astrobytes: hm, I have a TI graphical one somewhere but it needs a battery. Stuck with my 2 fx-83GT+
MSmits: ahh ok
MSmits: i had a TI graphical one from work. It died in a few weeks cuz the batteries leaked
MSmits: never got a new one
AntiSquid: geez just get some software with sci calc
MSmits: yeah, i dont need that shitty thing
Astrobytes: fx-83GT has some stats stuff but it's quicker to do it on the computer tbh
MSmits: you know, in my country we actually use graphical calculators on math exams
MSmits: it's mandatory curriculum
MSmits: i hate that, just let them use a plain calculator and draw their own graphs
Scarfield: its a good business for calculator producers
Astrobytes: Interesting. I was only allowed a basic non-graphing calculator without solving capabilities
MSmits: if you want to be fancy with tools, get a laptop and use real software
MSmits: Scarfield definitely is
MSmits: TI designed a calculator with buttons for physics constants
MSmits: like planck constant and stuff
MSmits: but physics commission made it illegal to use in exams, so we dont use it :(
MSmits: come on, who wouldnt want a light speed button
Astrobytes: Well, you are given all constants needed during the exam usually no
MSmits: they get a 200 page table book
Astrobytes: But yeah, the light speed button :D
MSmits: it's in table 7
Scarfield: i would, if it was a read one in my car
MSmits: but it's cooler when it's built in
Astrobytes: True
Marchete: "illegal to use in exams"ç
Marchete: like you need to know constants?
MSmits: it's too easy when they only press a button i guess
MSmits: they want to force students to use the table book thingy
MSmits: but I want my lightspeed button :(
Astrobytes: Sure. When I did my higher physics exam I had to actually recite the Planck length in 4 different ways
Zenoscave: "Smol"
Astrobytes: (I didn't)
MSmits: the planck length is something else. Also... recite, it's not a poem i dont think :P
Zenoscave: That's 1 way right?
Scarfield: lol zeno
Astrobytes: hah
Scarfield: making a poem out of it would help with memorizing it though
MSmits: probably, but why would anyone need the planck length
Astrobytes: recite it's numeric representation - better MSmits?
MSmits: dont you mean plancks constant?
Astrobytes: I pulled it out of the air for joking reasons!
Astrobytes: Jeez, never joke with physicists
MSmits: ohhh ok, well i take you seriously when you go quantum
Astrobytes: :D
MSmits: cuz you seem to know more than the average biologer
Scarfield: oof
MSmits: :P
Astrobytes: Yeah I do, it's true.
Marchete: I even google how to do a loop
MSmits: do a loop?
Astrobytes: I'll make a more educated joke next time :P
MSmits: quantum loop theory?
Marchete: for loop
MSmits: o
MSmits: I have a perfect solution for you Marchete
MSmits: use FP languages. No loops
Marchete: FP?
MSmits: functional programming
MSmits: haskell etc
Astrobytes: Forgettable Problems
Marchete: haskell? ugh
MSmits: well it's just one example
Astrobytes: Recursion recursion recursion recursion recursion...
MSmits: I wrote oware web app in elm :)
MSmits: got a 10/10 for it btw!
MSmits: maybe i mentioned that before
MSmits: don't care :P
MSmits: first time i am actually proud of anything i did during my studies. FP is hard to learn
Marchete: I prefer to learn from your bots rather than elm web apps
MSmits: sure, but this was not voluntary, i had to do it to pass the class :)
Zenoscave: Hey Marchete. How's your bot
Marchete: average
Marchete: it seems to win
AntiSquid: fix it
Zenoscave: ezpz
Marchete: ezpz
MSmits: Marchete is this a NN already or are you just using any old eval?
Zenoscave: Wait which bot are we talking about
MSmits: oware bot
Zenoscave: Gotcha
Marchete: just old eval from minimax, limited to -1..1
Astrobytes: Yes, I noticed it was improving Marchete
MSmits: maybe you just need a better eval then though
MSmits: it's not that easy to write a good one
Marchete: negamax was 15th, jacek around 20th
Astrobytes: In fact it's kinda painful
MSmits: the top 10, maybe 15 are pretty well fitted
Marchete: I think I have a bad simulation
Marchete: it's hard for me to know who wins when
MSmits: i think you underestimate the strength of the oware top players
Marchete: I need to reuse either referee or Agad_e arena code
MSmits: competition has been pretty fierce
Marchete: underestimate?
MSmits: yeah 20th is not weak
Astrobytes: It's a hardcore lb
Marchete: hmm
Marchete: 20th?
MSmits: there's a lot of guys with similar strengths
struct: top 20 ezpz
MSmits: between 5 and 15
MSmits: i think
Marchete: I don't think 20th+ are specially powerful
MSmits: no, i am not sure where the limit is
Marchete: my bot is negamax + some eval
MSmits: but top 10 is quite hard to reach
Marchete: like I tried two evals
Marchete: one mine and another from github
Marchete: and picked the best
Zenoscave: I think I finally have my C# Test-bed/Development policy ready for MAy
Astrobytes: Yeah, that's about right, maybe 5-10-11, I didn't look at scores recently
Marchete: I'd say top 15
MSmits: Marchete did you fit the values?
Zenoscave: Just need gitlab Devops now
MSmits: or did you just pick a few sets and kept the best?>
Marchete: only 2 sets
Marchete: one mine
Marchete: and another from internet
MSmits: ah well that makes a big difference
Marchete: no fitting
Astrobytes: Going in proper Zeno :)
MSmits: we spent weeks tuning these
Marchete: I won't tune that
Marchete: I'll put a NN to do the magic
MSmits: top 10 unreachable for a normal negamax with no tuning
Marchete: I need to tune CONSTANT_C
Zenoscave: Yup!
Marchete: I don't really now how to tune it
Zenoscave: to 440 Hz
Marchete: and this is a problem for training
MSmits: You don't even need to tune it if you use normal eval
Astrobytes: C is the least of your issues
MSmits: because you can set it to 1
MSmits: and fit all the others
Astrobytes: Isn't that A zeno
Marchete: what is my issue?
MSmits: then the C is a free parameter basically
Zenoscave: Oh yeah, whoops
Astrobytes: :grin:
YurkovAS: good eval example for oware
http://www.joansala.com/auale/strategy/en/
MSmits: with any set of parameters, you can set one to 1 or some other value and fit the others in relation
Marchete: I think I used that eval
MSmits: so just set c to 1
MSmits: I used that eval for my web app
Astrobytes: Yes as YurkovAS says it's a good starting point
MSmits: exactly that one
Marchete: <NNvalue[-1..1]> + ConstantC * whatever
Marchete: first is limited to -1 to 1
MSmits: but its wayyyy weaker than my live bots eval
Astrobytes: I can hardly read mine
Astrobytes: It's due a full rewrite/refactoring tbh
Zenoscave: Here's a question. How many of you try to keep coding standards and readability during a contest?
MSmits: try is the operative word
Astrobytes: Not as much as I do for multis
Marchete: it's exactly that heuristic
Marchete: https://github.com/joansalasoler/aalina/blob/7c7d0b27388c58b84edc2028277a71c3447329be/src/com/joansala/oware/OwareGame.java#L436-L464
MSmits: I *tr*
Marchete: what I used
MSmits: y
Zenoscave: I even try to handle Liskov stuff too
Zenoscave: SOLID etc.
MSmits: how many rollouts do you get Marchete?
Marchete: little
MSmits: also matters a lot
Marchete: because I need to store the gamestate
MSmits: well i store the gamestate on the node too
Astrobytes: DURING A CONTEST Zenoscave. I tip my hat
Marchete: I'd say 54k sim
MSmits: in turn 2 right?
Marchete: yes
MSmits: lemme do a test, sec
Zenoscave: Well I woudn't be a good TA for a class on SDLC's if I didn't Astro
Marchete: according to visits
MSmits: it's good to compare
MSmits: as soon as this ide loads, slow as hell
Astrobytes: Zeno: This is indeed, very true :)
MSmits: maybe because of the 25 kb opening book :P
Marchete: ....
Marchete: anyways I won't have a lot of simcount
Astrobytes: No, the Oware IDE takes ages to open
Marchete: I imagine the NN will take a lot of it
MSmits: seems about 220k visits
MSmits: but i expand all children, so thats about 4-5 visits
MSmits: each rollout
Marchete: I'd expect that you have like 800k or similar :D
Marchete: ahh
Marchete: 1M then
MSmits: no
MSmits: divide to get your result
MSmits: so only like 50k rollouts, but i eval all so its a bit more work
MSmits: counts as 1 visit for every pit
MSmits: this is why iget 220k
MSmits: so i dont think you're actually that slow
Astrobytes: I get 60-70k
Marchete: I'm probably counting badly
Astrobytes: But this version expands one at a time
MSmits: oh, that makes sense
MSmits: my eval is also quite heavy btw, it even has loops
MSmits: counting stuff over all pits and such
Marchete: do you priorize capture in some way on selection?
Astrobytes: Yeah I have a loop too
MSmits: selection is just based on UCT Marchete
Astrobytes: You can sure give it preference in your ewal
Astrobytes: or your eval
MSmits: http://chat.codingame.com/pastebin/372bfcb6-3d9c-4aa8-b801-535bea5a0fae
MSmits: its the same as every other bot basically
MSmits: i always do them like this, more or less
Marchete: what status mean?
Marchete: yeah
Marchete: seems similar
MSmits: solved or not
MSmits: -1 loss, 0 draw, 1 win, 2 ongoing
Marchete: hmm fastlogf(fastsqrtf
Marchete: no more lookup?
MSmits: nah, not because this is faster, but because i really hate the lookups
MSmits: this is a few % slower
Marchete: yeah I've had some problems
Marchete: when I have more visits than lookup
MSmits: i would artificially limit the rollouts to fix this
MSmits: and i dotn want to
MSmits: I think it might also hurt cache efficiency combined with other stuff
MSmits: like endgame books etc. Not sure, but it would make sense to me
MSmits: funny thing: When I went from 9 seeds to 10 seeds endgame book, my bot became worse
MSmits: 10 seeds db is so slow that it hurt my cache efficiency, is my theory
MSmits: i might put it back in when i write a NN
Marchete: I'm on the edge of start training
MSmits: NN does so few rollouts that the lookups dont hurt
Marchete: like it seems it's working
MSmits: nice one
Marchete: but I don't feel confident at all about it
MSmits: you've got stiff competition in the NN department also
MSmits: several different approaches
Astrobytes: ^
Marchete: not really
Marchete: I mean
Astrobytes: They're tough cookies
Marchete: I play solo
MSmits: not to mention nearly perfect play, so it's all down to 1 bad move out of 100 and you lose
Marchete: like I usually choose multis with low activity
Marchete: I'm trying to learn
MSmits: yeah its good for that
Astrobytes: I think we've noticed that Marchete :)
Marchete: and eventually put it some skeleton on github
Marchete: but not a ready to play bot
MSmits: just try not to judge too much by the top 5 strength, it's all really very deterministic
Marchete: don't worry
MSmits: except ja cek
MSmits: his bot is great
Astrobytes: And be confident, just get on with it
MSmits: it's all over the place and it's still strong
Marchete: I haaaate oware
Marchete: the whole simulation
Marchete: so I'm not for top 4
Marchete: or nothing
MSmits: I dont mind it, it's a bit finnicky
Marchete: just a working NN bot
MSmits: not sure finnicky is a word
MSmits: but it fits
Astrobytes: I thought it was really easy simulation, one or two tricky bits
Astrobytes: It is MSmits
MSmits: good
MSmits: there's more than one or two
MSmits: my sim has 4 tricky bits
Astrobytes: Might have one n, I've always used 2
MSmits: the first 60 were easy
Astrobytes: Yeah but you bitboarded it
Marchete: wtf is that
Marchete: https://github.com/MaximeCheramy/cg-oware/blob/master/src/main/java/com/codingame/game/Referee.java#L223-L230
MSmits: oh right, probably wasnt necessary to bb it
MSmits: since you also got a good simcount
Marchete: like you have an endgame before sim?
MSmits: I guess so
Astrobytes: If you can't move
Marchete: why does it takes all?
Marchete: if you can't move?
MSmits: it gets complicated when the last move at 200 does this
MSmits: do you get the seeds or not?
Marchete: it makes no sense at all
Astrobytes: That's the tricky bit
Marchete: and grand slam
Marchete: it's just tricky
MSmits: yeah it is Marchete
MSmits: tbh i dont even know if mine does everything perfectly
Astrobytes: Yes, it's kindof annoying to write but once it's done - it never needs to be touched again :)
Marchete: oware? I agree :D
Astrobytes: hahaha
MSmits: it's really the perfect game to practice nn's with once you do have the sim though
Marchete: that's why I'm here
Marchete: but I'm not sure about sim
Marchete: even the damn score is tricky
MSmits: ah well my sim is unintelligible, so i cant really help you
MSmits: if unintelligible is a word
Astrobytes: it is
MSmits: good
Marchete: un intelligible, two words :P
struct: I cant understand
Astrobytes: not in English :D
struct: If I change a static variable
struct: It doesnt show what it wants...
Marchete: I know
struct: I tried with a local variable and it worked
MSmits: again, try FP, no variables
Astrobytes: only immutable thingies
Marchete: haskell is the PCR/oware of languages
MSmits: yeah, it's mental
struct: Anyone has any idea why?
MSmits: eh, if you're starting to wonder what your variables really want, then you might need a break
Astrobytes: Hah! I think Haskell is quite elegant.
I am by no means what I'd consider proficient but it's pretty cool
Astrobytes: "It doesnt show what it wants..." What you want or what the variable wants?
Astrobytes: struct
MSmits: the resilience of a program coded in FP is amazing.
Marchete: yeah, if you know how to use it Haskell must be cool
struct: I think I know why
struct: I had a static array
Marchete: for non-FP
Marchete: it's hard
struct: That I was writing over his index
MSmits: tell me about it
struct: his max index
Astrobytes: Yeah I am very far from an FP expert
MSmits: for the first 2 sets of home work it took me almost a full day and was just as slow as the other students
MSmits: then you get used to it
MSmits: and you suddenly get to use all your coding skills again
struct: MSmits when can we expect a Yavalath paper?
MSmits: damn, there is so much to write up :(
Astrobytes: Well, when you have the time
MSmits: maybe I should just not try to come up with new stuff for a while and just do the things Marchete comes up with
Marchete: ok, smits says a C=1.0
Marchete: and these guys
Marchete: https://medium.com/oracledevs/lessons-from-alphazero-part-3-parameter-tweaking-4dceb78ed1e5
MSmits: then i have time to write up what i have
Marchete: put C=4.0
struct: Nah, you dont have to, I was just wondering, since you talked about it before.
MSmits: yes but Marchete, the C param is only meaninful in relation to your range of score
Marchete: range -1 to 1
Marchete: tanh
MSmits: even so, you have a st.dev, if your scores are in practice all between -0,02 and +0,02, it's different
MSmits: so you need to fit it
Marchete: I won't know how the NN will behave
MSmits: it is just 1 more param
Astrobytes: It's empirical MArchete
MSmits: I use 1.0 for two things
MSmits: when it's a loss/win thing like in uttt it is a good value
MSmits: so -1 to 1
MSmits: but also when i have all my eval params fitted in relation to c and i keep c at 1
MSmits: i did that in csb
MSmits: so my distance params are all super small
MSmits: 0.000001 and such
MSmits: in a nn it is going to be different
Marchete: I remember to reuse my ranges from xmas to PCR and they worked well
Marchete: like it's important to keep a good relationship
MSmits: yeah, you can always double your c and double all your eval params and you get the same bot
MSmits: it's all in relation
Astrobytes: Yes, as long as it's related
Marchete: ok, then I'd need to factor the NN output
Astrobytes: ah beat me again smits
Marchete: or adjust C
Astrobytes: yes
MSmits: thats it
MSmits: i would not worry so much about c
MSmits: if you pick it too low, all your nodes have the same visits
MSmits: no wait
MSmits: too high
MSmits: otherwise you will have 1 node with all visits, then c is too low
MSmits: somewhere between those two is ok
MSmits: worry about it when the nn starts getting better
Zenoscave: between all and neglible is ok
Marchete: I need to cache gamestates
Marchete: then I'll try to train
YurkovAS: Marchete i'm checked your eval function: 13-15 place with mcts ept depth 4, no solver, 25-30k rollouts second turn.
MSmits: at first i usually do a few tests in te ide, and if i see 100 visits, 100 visits 1000 visits 5k etc. It seems ok
Marchete: thanks yurkov
Marchete: then I need to improve the MCTS part
Marchete: negamax went like 15th with that
MSmits: i forgot to mention Marchete
Marchete: but my jacekmax won't go above 20th
MSmits: negamax did not do well on oware
MSmits: compared to mcts and similar
BlaiseEbuth: Fix your MCTS and go legend Marchete
Marchete: ezpz
MSmits: also Marchete, how come we dont hear from you for like a year and then you suddenly share all that brilliant stuff with us
Marchete: 30k rollouts is a lot
MSmits: first that NS monster and then that NN thigny
Marchete: I got tired and take some rest
MSmits: well it helped
Marchete: the NN is from months ago
Astrobytes: eh, my negamax was alright, my mcts+ept was way better. Doesn't tric still use ab nega?
Zenoscave: I think many of us needed a break for CG. I definitely did
MSmits: but thats tric
Marchete: NS was because xevi xevi got 700 with LAHC
MSmits: he is magical with that
Marchete: and I said wth
Astrobytes: It's not magic, I even got some tips from him after I helped him promote in Clobber
Marchete: in fact it was this kind of lucky shots
Marchete: after many trials
MSmits: well it's good to have you back again anyways
Marchete: with 12hrs solving time using 70vCPUs...
Marchete: I stalked a lot
Marchete: :eyes:
Astrobytes: Yeah, you used to pop up randomly at late European hours :)
Astrobytes: It is good to see you back
Astrobytes: (and all the other ones that seem to have returned)
Astrobytes: Why doesn't recar talk? I always see him online
Marchete: maybe in ru?
Astrobytes: hm, I don't spend much time in there, I always end up closing the chat window due to pms
Marchete: I have zero pm, but I didn't chat a lot because I haven't anything interesting to say
Marchete: except for NS, that I thought it was cool
Zenoscave: NS?
Marchete: number shifting
Zenoscave: negascout?
Zenoscave: ah.
Astrobytes: On the contrary Marchete, you clearly have lots of interesting things to say :)
MadKnight: #ru wants me to ban this guy who's been there for some time because he keeps spamming all kinds of random thoughts and his small updates for his coding project to the point where the entire chat is mostly filled with his junk, but for some reason i feel guilty.....
Marchete: I just wanted to reach 999 when the toad was online
MadKnight: Automaton2000 how do i fix me?
Automaton2000: i got to silver in uttt
Marchete: thx
struct: I also spam random thoughts
Marchete: so I just submitted when he was online and replying me :D
Marchete: damn toad and his damn puzzles
Zenoscave: I interject most of the time.
struct: Im just not spamming as much because I can't :zipper_mouth:
MadKnight: no struct it's so much worse with this guy
MadKnight: to the point where u get used to ignoring his messages by reflexes just like recus ignores Automaton2000
Automaton2000: in the pod racing game
Zenoscave: lol Which one Automaton2000?
Automaton2000: can i change my mind
Astrobytes: struct you always post your random thoughts
MadKnight: i mean i talked to this guy today and i felt like i got used to just ignoring what he says in such a short amount of time
struct: This time I cant
Marchete: me?
struct: its hard
Astrobytes: Ah right, ofc
Marchete: :*
Marchete: I'm going to sleep anyways, gn
Astrobytes: ConstantBolzmann MadKnight?
struct: gn
Astrobytes: gn Marchete
MSmits: gn, me too Marchete
MadKnight: ye, u gonna ban him right now?
MadKnight: lol
Astrobytes: Think I'm on my way too
struct: gn everyone
Astrobytes: gn all
BlaiseEbuth: Giant Noodle
Astrobytes: Tentacle
Astrobytes: bn satan :P
Marchete: is oware solved?
Astrobytes: not the abapa version
Astrobytes: It's complicated, ask again tomorrow
BlaiseEbuth: Big Noodle ?
Astrobytes: :grin:
MadKnight: wait who didn't go to sleep ?
RomanceDawn: you
MrFruit: slep god
MadKnight: RomanceDawn well so since u are a dawn...
MadKnight: RomanceDawn what is your pair of words supposed to mean ?
RomanceDawn: its
RomanceDawn: "the beging of a great journey "
RomanceDawn: its the title of the first chapter of ONE PIECE
MadKnight: oooh like a dawn of a great romance?
RomanceDawn: yeah yeah
MadKnight: dammit Automaton2000 i almost spelled "damn" instead of - dawn
Automaton2000: wish i had a few bugs
MadKnight: RomanceDammit
RomanceDawn: :grin:
MadKnight: everything must have an end RomanceDawn
MadKnight: so looking forward for the Dusk version
RomanceDawn: romanceDusk
MadKnight: Tusk
RomanceDawn: maybe the very last chapter of one piece
MadKnight: expecting it to be dark and ambient
MadKnight: i only like that stuff
MadKnight: not depressing
MadKnight: don't get those confused
RomanceDawn: gn mk
MadKnight: RomanceDawnWhenItsDusk
MadKnight: going to sleep until the dawn
MadKnight: Automaton2000 pls stop
Automaton2000: i don't see the point in the game
Zenoscave: Anid GW=50.00% [36.34% ,63.66% ] [W=50.00% L=50.00% D=0.00% ] [56]
kgautron GW=50.00% [36.34% ,63.66% ] [W=50.00% L=50.00% D=0.00% ] [56]
Dr. Dre GW=50.00% [37.02% ,62.98% ] [W=50.00% L=50.00% D=0.00% ] [62]
-- EVERYONE -- GW=50.00% [42.34% ,57.66% ] [W=50.00% L=50.00% D=0.00% ] [174 games]
Zenoscave: whoops sorry, thought that would pastebin
struct: which game Zenoscave?
Zenoscave: c4l
Zenoscave: I win exactly half my matches in against the next top players and the boss. I'm not sure if I should submit or wait
struct: you wont get pushed anytime soon
Zenoscave: true
Zenoscave: I'll tweak some params to match boss w/r and hope
MadKnight: Zenoscave pastebins are more than 4 lines of textg
MadKnight: u only had 4 lines
MadKnight: just remember that criteria
Zenoscave: Got it. thanks MK
MadKnight: Zenoscave anyway, what game is that ?
Zenoscave: Code4Life
MadKnight: and what is this data ?
Zenoscave: CG Benchmark
Zenoscave: Of my local bot
MadKnight: ooooh i need to re-solve this game
Zenoscave: You're wood still, yeah?
MadKnight: was almost legend during contest
MadKnight: i think it was my first contest where i couldn't reach legend
MadKnight: was sad
Zenoscave: WEPWEPWEP
MadKnight: need to reach legend everywhere now
kkkkl: gello
human37: your mom lol
jcrepes: my shirt do this
Upthea[n]ttic: that is my mom lol
MadKnight: guys guys calm down
MadKnight: this guy got a special channel for u guys
MadKnight: join it -> #hismom
MadKnight: struct are u working on any multi right now ?
struct: yes
MadKnight: oh, which one ?
struct: csb, im rewrinting my code
struct: Im doing avx for collisions
struct: instead of having 8x games with avx
MadKnight: oh i still need to make a super fancy template class that supports AVX natively
OBenjOne: if I want to output for debugging how to I stop the game from recognizing my print statement as a command to move?
OBenjOne: In Coders strike back*
struct: you print to stderr
struct: which language are you using?
OBenjOne: Python 3
Zenoscave: print(..., file=sys.stderr)
MadKnight: there's a comment telling u how to do that
Zenoscave: make sure you import sys
struct: print("Debug messages...", file=sys.stderr, flush=True)
MadKnight: u can find it
OBenjOne: where do I find that file to read it?
MadKnight: it's like a few lines abobe the main print()
Zenoscave: file just means where is it printing to
MadKnight: in a comment
MadKnight: it goes like
MadKnight:
- Write an answer using print
- To debug: print("Debug messages...", file=sys.stderr, flush=True)
OBenjOne: ok thanks Im sorry
MadKnight: don't worry
MadKnight: oh and some interesting info struct
MadKnight: so in c# there's actually a way to run scripts while the code is being parsed so that u would be getting raw tokens from the code
OBenjOne: hmmm maybe I should learn c++ then
MadKnight: and with all the cool c# features, u would be able to really beautify your c++ code by turning it in fancy c# code which just generates c++
MadKnight: OBenjOne wait why?
OBenjOne: Well it seems like just knowing python isn't quite enough in the real world
Zenoscave: Depends on what area you work in OBenjOne
MadKnight: it all depends on the kind of field u wanna work in
MadKnight: fancy maths or rocket science? well c++
MadKnight: some scripts for a server? well python
Zenoscave: Or Typescript
MadKnight: Zenoscript
Zenoscave: or Javascript. Or Java, or C#. There's tons of tools to use (A language is a tool) There's no "one tool to rule them all"
Zenoscave: just gotta learn what you like best and what's best for the job
OBenjOne: I don't exactly know what area I want to go into yet, and I think learning a variety of languages can't hurt
Zenoscave: ^ exactly
Zenoscave: but not Gloss over. Actually *learn* each one fairly well first
MadKnight: well just start with python and then switch to c#
MadKnight: that would be pretty much enough
OBenjOne: Ok thanks for the advice
MadKnight: c# is actually getting popular
Zenoscave: :thumbs_up:
Zenoscave: 18:33:13,718 INFO Final results : http://chat.codingame.com/pastebin/58dfbea2-a510-4ab6-bdc2-991aa307f7f3
Zenoscave: Looks possible worthwhile to submit?
Smelty: woAh 0-0
mybk: hello everybody, I was wondering, do you think CG profile is a good reference for recruiters to evaluate the candidate ?
PatrickMcGinnisII: Anyone set to receive professional emails ever get one? i would say, drop a linkedin or a github in there. *shrug*
MadKnight: my headphones driver stopped working for some reason
MadKnight: but not entirely
MadKnight: only the part controlling the mic
MadKnight: now the mic gives me raw sound without background sounds cleared
MadKnight: so now any soft and distant sound sounds very loud and i can't get them to stop going back to the headphones without turning down the mic entirely
Westicles: I suspect Russian hackers
MadKnight: dude i'm a russian
MadKnight: are u saying i'm such a bad russian i can't protect myself from russian hackers?
MadKnight: Automaton2000 stop getting triggered
Automaton2000: that's a great way to learn more about the game
MadKnight: anyway Westicles
Westicles: I don't want to point fingers, but policemen usually suspect the victim
MadKnight: what's time now for u ?
Westicles: Early evening
cViper971: how do i rank up in coders strike back
cViper971: im in first and ive beat the boss in the arena
cViper971: but im not ranking up
OBenjOne: It ranks you up after youve been ahead of the boss for a while
cViper971: ok
cViper971: do i need to stay in arena>
OBenjOne: At a certain time of day I think
OBenjOne: I dont actually know Im new too
cViper971: i can leave arena tho, right?
OBenjOne: yes I believe so
OBenjOne: My pod moved up In rank while I was coding a new one
OBenjOne: Quick question to anyone who's high rank in coders strike back: do you eventually get a current speed variable or should I code one myself by comparing distance and time?
Zenoscave: It changes the inputs later OBenjOne
OBenjOne: ok thanks then I won't waste time on that now
cViper971: anyone know when exactly it promotes
PatrickMcGinnisII: wow Astrobytes https://www.codingame.com/replay/541737806
Zenoscave: usually once an hour
cViper971: also how am i in first when i only did what the thing said
mybk: it should be displayed on top (I think)
OBenjOne: It gets harder
cViper971: like how come no one else out of thousands in my league can beat the boss
mybk: 1 hour to get from silver to gold league
cViper971: im in wood 1
cViper971: rn
mybk: isn't 5min or so
cViper971: idk why not being promoted then
mybk: I don't really remember
Zenoscave: cViper971 did you beat the boss?
cViper971: my first 2 promotions were like right after i beat boss
cViper971: now this one is taking some time
cViper971: yeah zeno
cViper971: in arena
Zenoscave: The wood leagues go faster
cViper971: the first 2 wood were like a minute
cViper971: this one is taking very long tho
Zenoscave: *Much* faster
cViper971: idk whats going on im still first
cViper971: been so for over 10 mins\
Zenoscave: Itll be about an hour
Zenoscave: should be a timer at the top if you refresh
cViper971: oooooooooof
cViper971: where
Zenoscave: what game
cViper971: coders strike back'
cViper971: the pod racing one
Zenoscave: you're still doing matches
Zenoscave: Your matches have to complete as well
Zenoscave: you're at 23%
Zenoscave: The timer will show up if you're above boss after all matches are done
cViper971: how do i see the matches i need to do
Zenoscave: click last battles
cViper971: ohh i see it
Zenoscave: it'll say a battles in progress and a %
cViper971: thx
Zenoscave: yw
PatrickMcGinnisII: Night of war is so wierd
PatrickMcGinnisII: 9th with a depth 1 eval
PatrickMcGinnisII: scared to up the depth
NotSureWhyThisWorks: Anyone know how to track time per move for c++? I am using the high_resolution_clock but it seems off by a bit
Westicles: #include <chrono>
Westicles: You can see it used here
Westicles: https://www.codingame.com/forum/t/collisions-in-fantastic-bits/2369/2
NotSureWhyThisWorks: I am new to C++ and that looks crazy lol
NotSureWhyThisWorks: Not sure if it is a glitch but it seems like the bot is timing out even though it is taking less than the time limit
Westicles: Yeah, sometimes you have to lower the limit until that goes away
NotSureWhyThisWorks: is that a thing that happens? or is my bot just dodgy
Westicles: I think last contest people had to set limits to 35ms intead of 50 to get it to work. But you might also have a bug, of course
NotSureWhyThisWorks: ok, I think it is probably both lol. I have to set to like 10ms
Smelty: o . o
cViper971: how do you make the best turns in coders strike back
cViper971: rn i have it if over 90 degrees thrust 0
cViper971: otherwise thrust 100
cViper971: but i was thinking having thrust = constant/degrees might be better
cViper971: its hard to tell tho
MadKnight: cViper971 still there ?
Smelty: wait how do we "solve" chess
Smelty: the weekly puzzle
davilla: write a program that never loses
ahthserhsluk: https://www.codingame.com/clashofcode/clash/17002497bb97d6d628f15b8cd3e3df6cfb50c36
