krugd/rustic_mcts
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

We propagate the reward for both sides

Browse Source 
The AI now properly choses the optimal path for the active player
This commit is contained in:
David Kruger 2025-06-27 16:07:33 -07:00
parent 0f9d4f0c4e
commit 6a33818238
1 changed files with 15 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
src/policy/backprop/mod.rs
View File

@ -69,18 +69,14 @@ fn standard_backprop<S: GameState>(
let mut current_id: usize = node_id; let mut current_id: usize = node_id;
loop { loop {
let node = arena.get_node_mut(current_id); let node = arena.get_node_mut(current_id);
let player = node.state.get_current_player().clone(); node.increment_visits();
match rewards.get(&player) { for (player, reward) in rewards.iter() {
Some(reward) => { node.record_player_reward(player.clone(), *reward);
node.increment_visits(); }
node.record_player_reward(player, *reward); if let Some(parent_id) = node.parent {
if let Some(parent_id) = node.parent { current_id = parent_id;
current_id = parent_id; } else {
} else { break;
break;
}
}
None => (),
} }
} }
} }
@ -94,19 +90,14 @@ fn weighted_backprop<S: GameState>(
let mut current_id: usize = node_id; let mut current_id: usize = node_id;
loop { loop {
let node = arena.get_node_mut(current_id); let node = arena.get_node_mut(current_id);
let player = node.state.get_current_player().clone();
let weight = weight_for_depth(depth_factor, node.depth); let weight = weight_for_depth(depth_factor, node.depth);
match rewards.get(&player) { for (player, reward) in rewards.iter() {
Some(reward) => { node.record_player_reward(player.clone(), (*reward) * weight);
node.increment_visits(); }
node.record_player_reward(player, (*reward) * weight); if let Some(parent_id) = node.parent {
if let Some(parent_id) = node.parent { current_id = parent_id;
current_id = parent_id; } else {
} else { break;
break;
}
}
None => (),
} }
} }
} }