🔒 how do i evaluate a connect 4 board position?

[topaz aspen]

i have a tablebase of 6000 positions and their relative evaluations, but outside of that, how can i make my own evaluator? and what do i need for it?

[vivid needle]

@topaz aspen

Python help channel opened

Remember to:

• Ask your Python question, not if you can ask or if there's an expert who can help.
• Show a code sample as text (rather than a screenshot) and the error message, if you've got one.
• Explain what you expect to happen and what actually happens.

:warning: Do not pip install anything that isn't related to your question, especially if asked to over DMs.

Closes after a period of inactivity, or when you send !close.

[topaz aspen]

any help is appreciated, might have late replies 🙏

[topaz aspen]

i’ve also heard of minimax and done some research

i just need a way to grade a board and i’ll be set

[topaz aspen]

anyone?

really?

[buoyant gulch]

well and how did that go?

[topaz aspen]

it went pretty well

i watched a sebastian lague video on it and i understand the tree searching stuff and recursion a little bit now

i just need a way to evaluate the board and i'll be set

but how tf do you evaluate a connect 4 board

[buoyant gulch]

I frankly don't have a clue myself, but have you heard about Code Bullet? he mostly does entertainment videos related to programming, but there are specs of technical stuff or, for example, where he has managed to copy the code from 😁
he happens to have a video on connect 4 (it's pretty far from "AI" despite the title...)
https://www.youtube.com/watch?v=XRVA5PMSKKE

[topaz aspen]

oh yeah i watched that and it's what inspired me to do this project, but i didnt understand the video in terms of technical stuff. thats why i wanted to come here

the tablebase got me 4 moves into the game 😭

[topaz aspen]

oh wait im dumb

i can just use the sources of what code bullet used

oh wait

i got the test cases from there 😭

[topaz aspen]

i still dont get how you evaluate a position though, thats the part im struggling with

[buoyant gulch]

I'm not entirely sure I will be able to help you with that right now, so uhh... yeah

[topaz aspen]

oh alr

who should i ask?

[buoyant gulch]

honestly all I can suggest is just wait for someone to show up and after some time maybe bump it in python general or if this gets automatically closed, open a new post

[topaz aspen]

alr will do

[desert peak]

you want to evaluate win conditions?

[topaz aspen]

yeah

so if i gave you a position like this:

hold on

[vivid needle]

:incoming_envelope: :ok_hand: applied timeout to @topaz aspen until <t:1713992114:f> (10 minutes) (reason: emoji spam - sent 42 emojis).

The <@&831776746206265384> have been alerted for review.

[desert peak]

how are you storing the board state? As a 2d array?

mods, do you mind unmuting? They're trying to demonstrate their board's state through emojis and tripped the bot

[mystic hare]

!unmute 566653183774949395

[vivid needle]

:incoming_envelope: :ok_hand: pardoned infraction timeout for @topaz aspen.

[mystic hare]

pls fewer emojis

[topaz aspen]

that was a fast 10 minutes

[topaz aspen]

okk

whats the limit?

[mystic hare]

... 25?

don't quote me though

[topaz aspen]

lets find out

[mystic hare]

just take a picture and send that though

[desert peak]

don't worry about showing me the board. I think I can help you regardless

is it a 2d list?

[topaz aspen]

yes

say i had a position like this

how would i evaluate it?

like how you can eval chess positions

[desert peak]

let's figure this out together

we should go top to bottom to avoid checking everything every single iteration

[topaz aspen]

(in case you dont know)

[topaz aspen]

yes

hold on brb just grabbing smth

[desert peak]

my first guess is to iterate through the outer list until we find a row that isnt empty. Then check the spaces around it.

[topaz aspen]

i saw someone saying you should check for chained pieces and evaluate it that way

[desert peak]

wait, we can optimize further

[topaz aspen]

how in the hell do you map out 8 directions for each of the 42 pieces

[desert peak]

imagine you have a board that's half full

assume that the last move wasn't a win

[topaz aspen]

[topaz aspen]

yes

[desert peak]

well, obviously if we make another move, the positions of the board that weren't near the new piece won't suddenly become wins

so we don't need to evaluate the entire board, just the positions near the new piece

(and so on, checking for a win condition)

like, hold on let me illustrate

[topaz aspen]

i got a half board but its kinda dodgy

[boreal summit]

This should give you some ideas to build of from

[topaz aspen]

my eyes already hurt

[boreal summit]

Just read it, it's not using any complicated terminology

[topaz aspen]

no like

small text

[naive orchid]

we are just trying to evaluate if the current board is in a win state right?

[boreal summit]

Ah, zoom function

[topaz aspen]

the only thing im stuck on is evaluating the board and giving it a score 😭

everything else is easy after that

[topaz aspen]

yep 👍

hold on brb

[desert peak]

right, say you put a piece in the last column. Obviously the pieces in the first column won't need to be evaluated again, because they've already been evaluated in the last move

[naive orchid]

also i just read about connect4 but correct me if i am wrong u basically win if u manage to connect 4 dots either horizontally, vertically, or diagonally

[desert peak]

so what you should do is:

1. place a piece
2. check the spaces immediately surrounding the placed piece (so left, down-left, down, down-right, right)
3. repeat step 2 for the piece next to it if the colour matches, and do so until until either there are 4 continuous pieces of the same colour in a straight line, or all "routes" are exhausted. Remember though, they must be in a straight line, so you should have a second "checking" function for straight lines, in addition to the initial proximity check

[boreal summit]

that is correct

[desert peak]

you're making me want to implement this

this could be a fun exercise

[topaz aspen]

yes

[topaz aspen]

ohh ok i think i see it now

[topaz aspen]

i smell DSA somewhere in here

[topaz aspen]

wait but how does this tie into evaluating the board? do i just count the number of 3s, 2s and just 1s with a relative score like 1 pt for each line of 1, 5 pts for each line of 2 and 10 pts for each line of 3 with infinity (or not at all) if theres 4 because thats a win anyway

[naive orchid]

do u have a few test cases? @topaz aspen

[topaz aspen]

uh, i had some test cases

you can get them here, lemme get a link

they dont have a file extension which is kinda odd

[desert peak]

whne you place a piece on the board, that's the only part of the board that changes. Parts of the board that are unaffected by the change won't magically become relevant to determining whether or not the win condition has been met, and you'll only know once you check. Since you have to start the check somewhere, the best place would be to check where the board saw a change (which is where the new piece was placed)

[topaz aspen]

http://blog.gamesolver.org/solving-connect-four/02-test-protocol/

[naive orchid]

ty

[topaz aspen]

they have the moves and then an eval

its supposed to calibrate your solver but idfk how to calibrate it 😭

[topaz aspen]

ah ok i see that. but my question still remains: how do you grade the board? how can i tell the difference (numerically in terms of score) between one board to another?

like with that board piece change, how can i tell if the position is better for the bot or for the opponent?

[vivid needle]

Python help channel closed

This help channel has been closed and it's no longer possible to send messages here. If your question wasn't answered, feel free to create a new post in #1035199133436354600. To maximize your chances of getting a response, check out this guide on asking good questions.