commit old stuff

src/elo.rs

@@ -9,7 +9,7 @@ use indicatif::{ProgressBar, ProgressStyle};
 use rand::seq::SliceRandom;
 use rayon::iter::{IntoParallelIterator, ParallelIterator};
 use skillratings::{
-    glicko2::{glicko2, Glicko2Rating},
+    glicko2::{confidence_interval, glicko2, Glicko2Rating},
     Outcomes, Rating,
 };
 use std::num::NonZero;
@@ -18,25 +18,19 @@ type AgentMaker = Box<dyn Fn(Piece) -> Box<dyn Agent>>;
 
 #[allow(dead_code)]
 pub fn run() {
-    let total_memory = 30_000_000_000;
+    let total_memory = 30_000_000_000; // 30 GB
     let num_threads = std::thread::available_parallelism()
         .map(NonZero::get)
         .expect("unable to get number of threads");
     let mem_per_thread = total_memory / num_threads;
 
     let fmv_base = FutureMoveConfig {
-        max_depth: 10,
-        min_arena_depth: 14,
-        top_k_children: 2,
-        up_to_minus: 10,
-        // max_arena_size: usize::MAX,
         max_arena_size: mem_per_thread / FutureMoves::ARENA_ENTRY_SIZE,
-        do_prune: false,
         print: false,
-        children_eval_method: Default::default(),
+        ..Default::default()
     };
 
-    let configs = [4, 5, 6]
+    let configs = [2, 3, 4, 5, 6, 7, 8]
         .into_iter()
         .map(move |d| FutureMoveConfig {
             max_depth: d,
@@ -127,7 +121,7 @@ pub fn run() {
         })
         .collect();
 
-    if true {
+    if false {
         vec.push((
             "RandomAgent".to_string(),
             Box::new(move |piece| Box::new(RandomAgent::new(piece))),
@@ -149,13 +143,22 @@ pub struct PlayerArena {
 impl std::fmt::Display for PlayerArena {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         let mut players_i: Vec<usize> = (0..self.players.len()).collect();
-        players_i.sort_by_key(|&i| -(self.players[i].2.rating() * 100.0) as i64);
+
+        players_i.sort_by(|&a, &b| {
+            self.players[b]
+                .2
+                .rating()
+                .total_cmp(&self.players[a].2.rating())
+        });
 
         for i in players_i {
+            let conf_interval = confidence_interval(&self.players[i].2);
+
             writeln!(
                 f,
-                "({:.2}): {}",
+                "({:.2}[+/-{:.2}]): {}",
                 self.players[i].2.rating(),
+                conf_interval.1 - self.players[i].2.rating(),
                 self.players[i].0
             )?;
         }
@@ -169,9 +172,8 @@ impl PlayerArena {
         Self {
             players: players
                 .into_iter()
-                .zip([Default::default()].into_iter().cycle())
-                // flatten tuple
-                .map(|((a, b), c)| (a, b, c))
+                // All starting ratings should be the default
+                .map(|(a, b)| (a, b, Default::default()))
                 .collect(),
         }
     }
@@ -236,7 +238,7 @@ impl PlayerArena {
             self.process_outcome(i, j, &o);
 
             if received_num > 0 {
-                term.clear_last_lines(self.players.len())
+                term.clear_last_lines(self.players.len() + 1)
                     .expect("unable to clear prev lines");
             }
             term.write_str(format!("{}", self).as_str())
@@ -244,8 +246,12 @@ impl PlayerArena {
 
             received_num += 1;
             p.inc(1);
+            // add extra newline after progressbar
+            println!();
+
             // break if all pairs were recieved
             if received_num == num {
+                drop(receiver);
                 break;
             }
         }
@@ -270,14 +276,12 @@ impl PlayerArena {
     }
 
     fn process_outcome(&mut self, player1: usize, player2: usize, outcome: &Outcomes) {
-        let (np1, np2) = glicko2(
+        (self.players[player1].2, self.players[player2].2) = glicko2(
             &self.players[player1].2,
             &self.players[player2].2,
             outcome,
             &Default::default(),
         );
-        self.players[player1].2 = np1;
-        self.players[player2].2 = np2;
     }
 
     fn play_two_inner(player_1: Box<dyn Agent>, player_2: Box<dyn Agent>) -> Outcomes {

src/logic/future_moves.rs

@@ -34,7 +34,7 @@ pub struct FutureMoves {
     board: Board,
 }
 
-#[derive(Copy, Clone, Allocative)]
+#[derive(Copy, Clone, Allocative, Default)]
 pub struct FutureMoveConfig {
     /// Max depth of that we should try and traverse
     pub max_depth: usize,
@@ -213,18 +213,7 @@ impl FutureMoves {
     }
 
     fn create_move(&self, coord: MoveCoord, board: Board, color: Piece) -> Move {
-        Move::new(
-            coord,
-            board,
-            color,
-            self.agent_color,
-            MoveValueConfig {
-                self_value_raw: matches!(
-                    self.config.children_eval_method,
-                    ChildrenEvalMethod::MinMaxProb
-                ),
-            },
-        )
+        Move::new(coord, board, color, self.agent_color, MoveValueConfig {})
     }
 
     fn generate_children_raw(&self, parent_idx: usize) -> Vec<Move> {
@@ -307,23 +296,6 @@ impl FutureMoves {
                     .iter()
                     .map(|&child| self.arena[child].value)
                     .collect::<Vec<_>>();
-
-                match self.config.children_eval_method {
-                    ChildrenEvalMethod::MinMax => {
-                        let child_value = if self.arena[idx].color == self.agent_color {
-                            // get best (for the adversary) enemy play
-                            // this assumes the adversary is playing optimally
-
-                            children_values.into_iter().map(|x| x.value).min()
-                        } else {
-                            children_values.into_iter().map(|x| x.value).max()
-                        }
-                        .unwrap_or(0);
-
-                        self.arena[idx].value.value =
-                            self.arena[idx].self_value.value + child_value;
-                    }
-                    ChildrenEvalMethod::MinMaxProb => {
                 let child_value = if self.arena[idx].color == self.agent_color {
                     // get best (for the adversary) enemy play
                     // this assumes the adversary is playing optimally
@@ -334,11 +306,18 @@ impl FutureMoves {
                 }
                 .cloned()
                 .unwrap_or(Default::default());
-
                 self.arena[idx].value = self.arena[idx].self_value;
+
+                match self.config.children_eval_method {
+                    ChildrenEvalMethod::MinMax => {
+                        self.arena[idx].value.value += child_value.value;
+                        self.arena[idx].value.set_state(child_value.state());
+                    }
+                    ChildrenEvalMethod::MinMaxProb => {
                         self.arena[idx]
                             .value
                             .populate_self_from_children(&children_values);
+
                         self.arena[idx].value.value += child_value.value;
                     }
                 }
@@ -389,7 +368,7 @@ impl FutureMoves {
                 ChildrenEvalMethod::MinMax => self.arena[x]
                     .children
                     .iter()
-                    .max_by_key(|&&idx| self.arena[idx].value.value),
+                    .max_by_key(|&&idx| self.arena[idx].value),
                 ChildrenEvalMethod::MinMaxProb => self.arena[x]
                     .children
                     .iter()

src/logic/move.rs

@@ -38,9 +38,7 @@ pub struct Move {
     pub is_trimmed: bool,
 }
 
-pub struct MoveValueConfig {
-    pub self_value_raw: bool,
-}
+pub struct MoveValueConfig {}
 
 impl Move {
     pub fn new(
@@ -48,7 +46,7 @@ impl Move {
         board: Board,
         color: Piece,
         agent_color: Piece,
-        mvc: MoveValueConfig,
+        _: MoveValueConfig,
     ) -> Self {
         let mut m = Move {
             coord,
@@ -76,30 +74,11 @@ impl Move {
             Winner::None => {}
         }
 
-        if mvc.self_value_raw {
         m.self_value.value =
             const { BoardValueMap::weighted() }.board_value(&board, agent_color) as i32;
-        } else {
-            m.self_value.value = m.compute_self_value(agent_color, &board, mvc) as i32;
-        }
         m
     }
 
-    fn compute_self_value(&self, agent_color: Piece, board: &Board, _mvc: MoveValueConfig) -> i16 {
-        if self.winner == Winner::Player(!agent_color) {
-            // if this board results in the opponent winning, MAJORLY negatively weigh this move
-            // NOTE! this branch isn't completely deleted because if so, the bot wouldn't make a move.
-            // We shouldn't prune branches because we still need to always react to the opponent's moves
-            return i16::MIN + 1;
-        } else if self.winner == Winner::Player(agent_color) {
-            // results in a win for the agent
-            return i16::MAX - 1;
-        }
-
-        // I guess ignore Ties here, don't give them an explicit value,
-        const { BoardValueMap::weighted() }.board_value(board, agent_color)
-    }
-
     /// Sort children of the [`Move`] by their self_value in `arena`
     pub fn sort_children(&mut self, arena: &[Move]) {
         self.children.sort_by(|&a, &b| {

src/logic/mvs.rs

@@ -56,19 +56,19 @@ impl MoveValueStats {
         self.state = state;
     }
 
+    pub const fn state(&self) -> Option<MVSGameState> {
+        self.state
+    }
+
     pub fn populate_self_from_children(&mut self, others: &[Self]) {
-        self.wins = others
-            .iter()
-            .map(|x| x.wins + (x.state == Some(MVSGameState::Win)) as u16)
-            .sum::<u16>();
-        self.losses = others
-            .iter()
-            .map(|x| x.losses + (x.state == Some(MVSGameState::Loss)) as u16)
-            .sum::<u16>();
-        self.ties = others
-            .iter()
-            .map(|x| x.ties + (x.state == Some(MVSGameState::Tie)) as u16)
-            .sum::<u16>();
+        (self.wins, self.losses, self.ties) =
+            others.iter().fold((0, 0, 0), |(wins, losses, ties), x| {
+                (
+                    wins + x.wins + (x.state == Some(MVSGameState::Win)) as u16,
+                    losses + x.losses + (x.state == Some(MVSGameState::Loss)) as u16,
+                    ties + x.ties + (x.state == Some(MVSGameState::Tie)) as u16,
+                )
+            });
     }
 }
 
@@ -85,7 +85,7 @@ impl Ord for MoveValueStats {
         }
 
         let (s_cw, o_cw) = (self.chance_win(), other.chance_win());
-        if s_cw.is_some() | o_cw.is_some() {
+        if s_cw.is_some() || o_cw.is_some() {
             if s_cw > o_cw {
                 return Ordering::Greater;
             } else if o_cw > s_cw {

src/repr/pos_map.rs

@@ -14,12 +14,12 @@ impl<T: Copy> PosMap<T> {
         Self(MaybeUninit::zeroed().assume_init())
     }
 
-    pub const fn from(v: [[T; Board::SIZE as usize]; Board::SIZE as usize]) -> Self {
+    pub const fn from(mut v: [[T; Board::SIZE as usize]; Board::SIZE as usize]) -> Self {
         let mut n = unsafe { Self::uninit() };
 
         const_for!(i in 0..Board::SIZE => {
             const_for!(j in 0..Board::SIZE => {
-                n.set(CoordPair::from_axes(i, j), v[i as usize][j as usize]);
+                std::mem::swap(n.get_mut(CoordPair::from_axes(i, j)), &mut v[i as usize][j as usize]);
             });
         });
         n