nelfin / othello Goto Github PK

Just noting this idea down.

New learnable feature for a player: state-cut lower bound.

Perform simple static evaluations during minimax and cut out trees where we see a VERY poor state.

ADVANTAGES:

• Dunno yet

PROBLEMS:

• Potential huge overhead in minimax
• Potential to lose some pretty a-ok states.

Should the slides be a total bare-bone outline or include some level of
explanation?

Oh, also:
One click per slide or one click per dot point?

Personally I kinda like one click per dot point. Stops people from reading ahead.

Just reminding myself, to fix bugs in 7b0c355

Wikipedia Negamax

Basically uses max(a, b) = -min(-a, -b)

function negamax(node, depth, α, β, color)
    if node is a terminal node or depth = 0
        return color * the heuristic value of node
    else
        foreach child of node
            val := -negamax(child, depth-1, -β, -α, -color)
            {the following if statement constitutes alpha-beta pruning}
            if val≥β
                return val
            if val≥α
                α:=val
        return α

Where could they go? I'm currently trying to work on simple move ordering to help alpha-beta cutoffs.

Needs minimax and alpha-beta to begin with.

Eclipse + Git is not so bad, but I'm kinda fumbling my way through the integration/folder layout bit.

Minimax seems to be losing more than it should against a random player, about 15-20% loss (minimal draws, win % was around 80%).

A meta player could use lots of different plays to be amazing and awesome at Othello and own everyone in the tournament.

Figured we should get a nice early start on the report as soon as possible.

I volunteer! I'll probably start work on it during the week.

I've seen my minimax lose against a greedy player, when playing first and second. Sometimes it loses so badly that the board doesn't even get filled: the greedy player is able to eliminate all of the minimax player's pieces.

This makes me think that it's playing deliberately badly, but flipping the sign of the values returned didn't seem to have any effect. I'll try implementing it again, but I'm on my third attempt now. I think I'll have to modify the server to make the board smaller and do it by hand.

Another random thought that probably is relevant to no-one.

Was wondering how hard-coded features need to be; and considering the possibility of
a Feature Markup Language (FML). Seems like it would be a lot of pointless work though... really only wanna do it for the acronym.

Because we're supposedly professional and whatnot.

This could be done with SGF

Sorry for being MIA so long; but SML finally finished!

I've spent today having a browse through all the code and I gotta say it's looking mighty fine.

Is there anything important left to do before test runs tomorrow?
I could look at doing some code cleaning? (we're supposed to email code to Stephen before the meeting)

A priority list for next tuesday. Please comment and add stuff.

MUST HAVERS:

1. More features
2. Meta player enhancements
   • Time awareness
   • Game stage awareness
3. Opening book player
4. Even more features

NICE TO HAVERS:

1. Not sure/Forgot

I can't figure out how to compile the code to steal all your sekrits. Serious issue.

Also I am a gay.

Tomorrow, I have 2 exams and I'm going to be busy pretty much all day. If you guys can think of anything else I should do on the report, best tell me now, I wont be able to do much tomorrow.

Make everything tidy and presentable

I'll get onto this for now. Looks like it'll be a fun holidteaching break.

Also, Josh is not the top-scorer for this repository:

http://github-high-scores.heroku.com/nelfin/othello/high_scores/

No way of evaluating an intermediate state (not that it even has any "intermediate states" at this stage).

Sorry guys but I am gonna be really busy with 4670 over the break... probably can't do much this first week;
but next week I'll try split my time evenly between the two.