Help please with alpha-beta

[Alexandr Novikov]

Hi friends! I am a beginner chess programmer , this is my hobby, so do not judge strictly. I wrote a simple chess program with graphics interface(pascal abc net). It's five ply or depth (I do not understand the difference between them) search with minimax method. My search procedure without reduction and my code looks like this:
//--------------------------------------------------------------------------------------
procedure search;
var .....
black_move_generator;
for i:=0 to bcnt - 1; //all possible black moves
make_move;
first_ply; //call next iteration
if i = 0 then best_score := best_score1 else
if best_score1 < best_score then best_score := best_score1; //min
unmake_move;
//--------------------------------------------------------------------------------------
procedure two_ply;
var .....
white_move_generator;
for i:=0 to wcnt - 1; //all possible white moves
make_move;
tree_ply; //call next iteration
if i = 0 then best_score1 := best_score2 else
if best_score2 > best_score1 then best_score1 := best_score2; //max
unmake_move;
//-------------------------------------------------------------------------------------- procedure tree_ply;
var .....
black_move_generator;
for i:=0 to bcnt - 1; //all possible black moves
make_move;
four_ply; //call next iteration
if i = 0 then best_score2 := best_score3 else
if best_score3 < best_score2 then best_score2 := best_score3; // min
unmake_move;
//--------------------------------------------------------------------------------------procedure four_ply;
var .....
white_move_generator;
for i:=0 to wcnt - 1; //all possible white moves
make_move;
five_ply; //call next iteration
if i = 0 then best_score3 := best_score4 else
if best_score4 > best_score3 then best_score3 := best_score4; //max
unmake_move;
//--------------------------------------------------------------------------------------procedure five_ply;
var .....
black_move_generator;
for i:=0 to bcnt - 1; //all possible black moves
make_move;
evaluate; //call evaluate function
if i = 0 then best_score4 := eval else
if eval < best_score4 then best_score4 := eval; // min
unmake_move;
//--------------------------------------------------------------------------------------
Minimax works fine, but I do not understand both should work alpha beta( Please, explain me, how on my pseudo code should work alpha beta. I know, that for a-b very important moves ordering and i have simple moves ordering.
thank you in advance.
best
Alexandr Novikov
p.s.I do not want to read other people code, I want to understand everything myself

[Henk]

First do you understand this ?

upperbound(white to move ) = -lowerbound(black to move)
value( white to move) = - value (black to move )


I use something like this:

Code: Select all

search (lb, ub)
{
   val = -infinite;
    for each possible move
    {
       make move -> changes also side to move
        val = max (val, -search(-ub, -lb));
       undo  move -> changes also side to move
    }
     return val
}

[Alexandr Novikov]

First do you understand this ?

upperbound(white to move ) = -lowerbound(black to move)
value( white to move) = - value (black to move )

minimax? yes?

[Henk]

See second part of previous pos. I think you can find the difference yourself.

[Alexandr Novikov]

See second part of previous pos. I think you can find the difference yourself.

ok, but where alpha beta in your code? sorry, I don't understand

[AlvaroBegue]

You are using procedures, but the main thing each subroutine is doing is computing a value (e.g., `best_score2'). Why not write functions and return the best score? That's how most of us do it.

You don't need different functions for different levels in the tree: You have a single function that takes an argument that says how much depth is left.

I recommend you use the negamax convention, where scores are positive for the side to move, and then you also don't need separate code for max nodes and min nodes.

So, ignoring alpha-beta pruning for the time being, can you rewrite your code following my advice above? Adding alpha-beta pruning to that code is a small modification (although it's not particularly easy to understand).

[Henk]

See second part of previous pos. I think you can find the difference yourself.

ok, but where alpha beta in your code? sorry, I don't understand

search(lb, ub) is my interpretation of search(alpha, beta)

[bob]

First do you understand this ?

upperbound(white to move ) = -lowerbound(black to move)
value( white to move) = - value (black to move )


I use something like this:

Code: Select all

search (lb, ub)
{
   val = -infinite;
    for each possible move
    {
       make move -> changes also side to move
        val = max (val, -search(-ub, -lb));
       undo  move -> changes also side to move
    }
     return val
}

So NO alpha/beta at all? Pure minimax???

[Henk]

First do you understand this ?

upperbound(white to move ) = -lowerbound(black to move)
value( white to move) = - value (black to move )


I use something like this:

Code: Select all

search (lb, ub)
{
   val = -infinite;
    for each possible move
    {
       make move -> changes also side to move
        val = max (val, -search(-ub, -lb));
       undo  move -> changes also side to move
    }
     return val
}

So NO alpha/beta at all? Pure minimax???

Fortunately I said I use something like this. I am a bad tutor. I have no patience.

[Alexandr Novikov]

First do you understand this ?

upperbound(white to move ) = -lowerbound(black to move)
value( white to move) = - value (black to move )


I use something like this:

Code: Select all

search (lb, ub)
{
   val = -infinite;
    for each possible move
    {
       make move -> changes also side to move
        val = max (val, -search(-ub, -lb));
       undo  move -> changes also side to move
    }
     return val
}

So NO alpha/beta at all? Pure minimax???

I thought so too