feat: Exhaustive search
Browse files- README.md +1 -0
- python/benchmark.py +8 -4
- python/othello/ui.py +1 -1
- src/ai.rs +20 -8
README.md
CHANGED
|
@@ -21,5 +21,6 @@ othello-ui
|
|
| 21 |
|
| 22 |
```bash
|
| 23 |
uv sync
|
|
|
|
| 24 |
python python/othello/ui.py
|
| 25 |
```
|
|
|
|
| 21 |
|
| 22 |
```bash
|
| 23 |
uv sync
|
| 24 |
+
maturin develop
|
| 25 |
python python/othello/ui.py
|
| 26 |
```
|
python/benchmark.py
CHANGED
|
@@ -23,16 +23,20 @@ def run_matches(bot1, bot2, n):
|
|
| 23 |
return cnts
|
| 24 |
|
| 25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
if __name__ == "__main__":
|
| 27 |
for i in range(2, 10):
|
| 28 |
-
bot1 = AlphaBetaBot(i)
|
| 29 |
-
bot2 = AlphaBetaBot(i + 1)
|
| 30 |
-
print(f"
|
| 31 |
draw, win, lost = run_matches(bot1, bot2, 10)
|
| 32 |
print(f"Win: {win} | Draw: {draw} | Lost: {lost}")
|
| 33 |
print("----")
|
| 34 |
|
| 35 |
-
print(f"
|
| 36 |
draw, win, lost = run_matches(bot2, bot1, 10)
|
| 37 |
print(f"Win: {win} | Draw: {draw} | Lost: {lost}")
|
| 38 |
print("----")
|
|
|
|
| 23 |
return cnts
|
| 24 |
|
| 25 |
|
| 26 |
+
def _str(bot):
|
| 27 |
+
return f"AlphaBetaBot({bot.depth}, {bot.exhaustive_depth})"
|
| 28 |
+
|
| 29 |
+
|
| 30 |
if __name__ == "__main__":
|
| 31 |
for i in range(2, 10):
|
| 32 |
+
bot1 = AlphaBetaBot(i, 14)
|
| 33 |
+
bot2 = AlphaBetaBot(i + 1, 14)
|
| 34 |
+
print(f"{_str(bot1)} vs {_str(bot2)})")
|
| 35 |
draw, win, lost = run_matches(bot1, bot2, 10)
|
| 36 |
print(f"Win: {win} | Draw: {draw} | Lost: {lost}")
|
| 37 |
print("----")
|
| 38 |
|
| 39 |
+
print(f"{_str(bot2)} vs {_str(bot1)})")
|
| 40 |
draw, win, lost = run_matches(bot2, bot1, 10)
|
| 41 |
print(f"Win: {win} | Draw: {draw} | Lost: {lost}")
|
| 42 |
print("----")
|
python/othello/ui.py
CHANGED
|
@@ -20,7 +20,7 @@ app, rt = fast_app(
|
|
| 20 |
)
|
| 21 |
|
| 22 |
games = {}
|
| 23 |
-
bot = AlphaBetaBot(
|
| 24 |
|
| 25 |
|
| 26 |
@rt("/")
|
|
|
|
| 20 |
)
|
| 21 |
|
| 22 |
games = {}
|
| 23 |
+
bot = AlphaBetaBot(7, 15)
|
| 24 |
|
| 25 |
|
| 26 |
@rt("/")
|
src/ai.rs
CHANGED
|
@@ -12,16 +12,25 @@ pub trait Bot {
|
|
| 12 |
pub struct AlphaBetaBot {
|
| 13 |
#[pyo3(get)]
|
| 14 |
depth: usize,
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
}
|
| 16 |
|
| 17 |
impl Bot for AlphaBetaBot {
|
| 18 |
fn find_move(&self, g: &Game) -> i32 {
|
| 19 |
let count = (g.board.0 + g.board.1).count_ones() as usize;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
let (_, move_) = self.do_search(
|
| 21 |
&AlphaBetaEval { count },
|
| 22 |
&mut rand::thread_rng(),
|
| 23 |
&g.board,
|
| 24 |
-
|
| 25 |
-i32::MAX,
|
| 26 |
i32::MAX,
|
| 27 |
);
|
|
@@ -32,8 +41,11 @@ impl Bot for AlphaBetaBot {
|
|
| 32 |
#[pymethods]
|
| 33 |
impl AlphaBetaBot {
|
| 34 |
#[new]
|
| 35 |
-
pub fn new(depth: usize) -> Self {
|
| 36 |
-
Self {
|
|
|
|
|
|
|
|
|
|
| 37 |
}
|
| 38 |
|
| 39 |
#[pyo3(name = "find_move")]
|
|
@@ -60,7 +72,7 @@ impl AlphaBetaBot {
|
|
| 60 |
let board = board.pass_move();
|
| 61 |
let moves = board.available_moves();
|
| 62 |
if moves == 0 {
|
| 63 |
-
return (eval.final_evaluate(&board), -1);
|
| 64 |
}
|
| 65 |
let (score, _) = self.do_search(eval, rng, &board, depth - 1, -beta, -alpha);
|
| 66 |
return (-score, -1);
|
|
@@ -106,7 +118,7 @@ impl AlphaBetaEval {
|
|
| 106 |
let mut score = scorer(board.0) - scorer(board.1) + 10 * n_moves0 - 10 * n_moves1;
|
| 107 |
if self.count > 54 {
|
| 108 |
let (cnt0, cnt1) = board.count();
|
| 109 |
-
score += 2 * (self.count as i32 - 54) * (cnt0 - cnt1) as i32
|
| 110 |
}
|
| 111 |
score
|
| 112 |
}
|
|
@@ -114,9 +126,9 @@ impl AlphaBetaEval {
|
|
| 114 |
fn final_evaluate(&self, board: &BitBoard) -> i32 {
|
| 115 |
let (sc1, sc2) = board.count();
|
| 116 |
if sc1 < sc2 {
|
| 117 |
-
-i32::MAX
|
| 118 |
} else if sc1 > sc2 {
|
| 119 |
-
i32::MAX
|
| 120 |
} else {
|
| 121 |
0
|
| 122 |
}
|
|
@@ -144,7 +156,7 @@ mod tests {
|
|
| 144 |
#[test]
|
| 145 |
fn test() {
|
| 146 |
let mut b = Game::default();
|
| 147 |
-
let ai = [AlphaBetaBot
|
| 148 |
|
| 149 |
while !b.state.ended {
|
| 150 |
let pos = ai[b.current_player].find_move(&b);
|
|
|
|
| 12 |
pub struct AlphaBetaBot {
|
| 13 |
#[pyo3(get)]
|
| 14 |
depth: usize,
|
| 15 |
+
|
| 16 |
+
// Do exhaustive search when there is less than exhaustive_depth empty square
|
| 17 |
+
#[pyo3(get)]
|
| 18 |
+
exhaustive_depth: usize,
|
| 19 |
}
|
| 20 |
|
| 21 |
impl Bot for AlphaBetaBot {
|
| 22 |
fn find_move(&self, g: &Game) -> i32 {
|
| 23 |
let count = (g.board.0 + g.board.1).count_ones() as usize;
|
| 24 |
+
let depth = if count > 64 - self.exhaustive_depth {
|
| 25 |
+
100
|
| 26 |
+
} else {
|
| 27 |
+
self.depth
|
| 28 |
+
};
|
| 29 |
let (_, move_) = self.do_search(
|
| 30 |
&AlphaBetaEval { count },
|
| 31 |
&mut rand::thread_rng(),
|
| 32 |
&g.board,
|
| 33 |
+
depth,
|
| 34 |
-i32::MAX,
|
| 35 |
i32::MAX,
|
| 36 |
);
|
|
|
|
| 41 |
#[pymethods]
|
| 42 |
impl AlphaBetaBot {
|
| 43 |
#[new]
|
| 44 |
+
pub fn new(depth: usize, exhaustive_depth: usize) -> Self {
|
| 45 |
+
Self {
|
| 46 |
+
depth,
|
| 47 |
+
exhaustive_depth,
|
| 48 |
+
}
|
| 49 |
}
|
| 50 |
|
| 51 |
#[pyo3(name = "find_move")]
|
|
|
|
| 72 |
let board = board.pass_move();
|
| 73 |
let moves = board.available_moves();
|
| 74 |
if moves == 0 {
|
| 75 |
+
return (-eval.final_evaluate(&board), -1);
|
| 76 |
}
|
| 77 |
let (score, _) = self.do_search(eval, rng, &board, depth - 1, -beta, -alpha);
|
| 78 |
return (-score, -1);
|
|
|
|
| 118 |
let mut score = scorer(board.0) - scorer(board.1) + 10 * n_moves0 - 10 * n_moves1;
|
| 119 |
if self.count > 54 {
|
| 120 |
let (cnt0, cnt1) = board.count();
|
| 121 |
+
score += 2 * (self.count as i32 - 54) * (cnt0 - cnt1) as i32
|
| 122 |
}
|
| 123 |
score
|
| 124 |
}
|
|
|
|
| 126 |
fn final_evaluate(&self, board: &BitBoard) -> i32 {
|
| 127 |
let (sc1, sc2) = board.count();
|
| 128 |
if sc1 < sc2 {
|
| 129 |
+
-i32::MAX + (64 + sc1 - sc2)
|
| 130 |
} else if sc1 > sc2 {
|
| 131 |
+
i32::MAX - (64 + sc1 - sc2)
|
| 132 |
} else {
|
| 133 |
0
|
| 134 |
}
|
|
|
|
| 156 |
#[test]
|
| 157 |
fn test() {
|
| 158 |
let mut b = Game::default();
|
| 159 |
+
let ai = [AlphaBetaBot::new(3, 0), AlphaBetaBot::new(6, 10)];
|
| 160 |
|
| 161 |
while !b.state.ended {
|
| 162 |
let pos = ai[b.current_player].find_move(&b);
|