probabilistic progress

src/elo.rs

@@ -29,7 +29,7 @@ pub fn run() {
         children_eval_method: Default::default(),
     };
 
-    let configs = [4, 5, 6]
+    let configs = [6]
         .into_iter()
         .map(move |d| FutureMoveConfig {
             max_depth: d,
@@ -119,14 +119,17 @@ pub fn run() {
             )
         })
         .collect();
-    vec.push((
+
-        "RandomAgent".to_string(),
+    if false {
-        Box::new(move |piece| Box::new(RandomAgent::new(piece))),
+        vec.push((
-    ));
+            "RandomAgent".to_string(),
+            Box::new(move |piece| Box::new(RandomAgent::new(piece))),
+        ));
+    }
 
     let mut arena = PlayerArena::new(vec);
 
-    arena.prop_arena(1000);
+    arena.prop_arena(100);
 
     println!("{}", arena);
 }
@@ -172,7 +175,10 @@ impl PlayerArena {
             .map(|&(i, j)| {
                 (
                     (i, j),
-                    Self::create_agents(&self.players[i].1, &self.players[j].1),
+                    (
+                        (&self.players[i].1)(Piece::Black),
+                        (&self.players[j].1)(Piece::White),
+                    ),
                 )
             })
             .collect::<Vec<_>>();
@@ -221,14 +227,16 @@ impl PlayerArena {
 
         while let Ok((i, j, o)) = receiver.recv() {
             self.process_outcome(i, j, &o);
-            received_num += 1;
-            p.inc(1);
 
-            term.clear_last_lines(self.players.len())
+            if received_num > 0 {
-                .expect("unable to clear prev lines");
+                term.clear_last_lines(self.players.len())
+                    .expect("unable to clear prev lines");
+            }
             term.write_str(format!("{}", self).as_str())
                 .expect("unable to write leaderboard");
 
+            received_num += 1;
+            p.inc(1);
             // break if all pairs were recieved
             if received_num == num {
                 break;
@@ -265,19 +273,13 @@ impl PlayerArena {
         self.players[player2].2 = np2;
     }
 
-    fn create_agents(
-        player_1_fn: &AgentMaker,
-        player_2_fn: &AgentMaker,
-    ) -> (Box<dyn Agent>, Box<dyn Agent>) {
-        (player_1_fn(Piece::Black), player_2_fn(Piece::White))
-    }
-
     fn play_two_inner(player_1: Box<dyn Agent>, player_2: Box<dyn Agent>) -> Outcomes {
         let result = GameInner::new(
             player_1,
             player_2,
             false,
-            Board::random(rand::random_range(1..=15)),
+            // Board::random(rand::random_range(20..=30)),
+            Board::STARTING_POSITION,
         )
         .expect("unable to create game")
         .loop_until_result();

src/logic/future_moves.rs

@@ -315,23 +315,14 @@ impl FutureMoves {
                             // get best (for the adversary) enemy play
                             // this assumes the adversary is playing optimally
 
-                            children_values
+                            children_values.into_iter().map(|x| x.value).min()
-                                .into_iter()
-                                .min_by_key(|x| x.value)
-                                .map(|x| x.value)
                         } else {
-                            children_values
+                            children_values.into_iter().map(|x| x.value).max()
-                                .into_iter()
-                                .max_by_key(|x| x.value)
-                                .map(|x| x.value)
                         }
                         .unwrap_or(0);
-                        // we use `depth` and divided `self_value` by it, idk if this is worth it
+
-                        // we should really setup some sort of ELO rating for each commit, playing them against
-                        // each other or something, could be cool to benchmark these more subjective things, not
-                        // just performance (cycles/time wise)
                         self.arena[idx].value.value =
-                            self.arena[idx].self_value as i32 + child_value;
+                            self.arena[idx].self_value.value as i32 + child_value;
                     }
                     ChildrenEvalMethod::MinMaxProb => {
                         let child_value = if self.arena[idx].color == self.agent_color {
@@ -344,16 +335,12 @@ impl FutureMoves {
                         }
                         .cloned()
                         .unwrap_or(Default::default());
-                        // we use `depth` and divided `self_value` by it, idk if this is worth it
+
-                        // we should really setup some sort of ELO rating for each commit, playing them against
+                        self.arena[idx].value = self.arena[idx].self_value;
-                        // each other or something, could be cool to benchmark these more subjective things, not
-                        // just performance (cycles/time wise)
                         self.arena[idx]
                             .value
                             .populate_self_from_children(&children_values);
-
+                        self.arena[idx].value.value += child_value.value;
-                        self.arena[idx].value.value =
-                            self.arena[idx].self_value as i32 + child_value.value;
                     }
                 }
             }

src/logic/move.rs

@@ -29,7 +29,7 @@ pub struct Move {
     pub value: MoveValueStats,
 
     /// What is the inherit value of this move (not including children)
-    pub self_value: i16,
+    pub self_value: MoveValueStats,
 
     /// Which color made a move on this move?
     pub color: Piece,
@@ -58,28 +58,29 @@ impl Move {
             value: Default::default(),
             color,
             is_trimmed: false,
-            self_value: 0,
+            self_value: Default::default(),
         };
 
         // set wins/losses values appropriately
         match m.winner {
             Winner::Player(piece) => {
                 if piece == agent_color {
-                    m.value.set_state(Some(MVSGameState::Win));
+                    m.self_value.set_state(Some(MVSGameState::Win));
                 } else {
-                    m.value.set_state(Some(MVSGameState::Loss));
+                    m.self_value.set_state(Some(MVSGameState::Loss));
                 }
             }
             Winner::Tie => {
-                m.value.set_state(Some(MVSGameState::Tie));
+                m.self_value.set_state(Some(MVSGameState::Tie));
             }
             Winner::None => {}
         }
 
         if mvc.self_value_raw {
-            m.self_value = const { BoardValueMap::weighted() }.board_value(&board, agent_color);
+            m.self_value.value =
+                const { BoardValueMap::weighted() }.board_value(&board, agent_color) as i32;
         } else {
-            m.self_value = m.compute_self_value(agent_color, &board, mvc);
+            m.self_value.value = m.compute_self_value(agent_color, &board, mvc) as i32;
         }
         m
     }

src/logic/mvs.rs

@@ -8,48 +8,44 @@ pub enum MVSGameState {
     Loss = -1,
 }
 
-#[derive(Clone, Copy, Debug, Allocative, PartialEq, Eq, Default)]
+#[derive(Clone, Copy, Debug, Allocative, PartialEq, Eq, Default, PartialOrd)]
 pub struct MoveValueStats {
     state: Option<MVSGameState>,
     wins: u16,
     losses: u16,
+    ties: u16,
     pub value: i32,
 }
 
 impl MoveValueStats {
     #[cfg(test)]
-    pub const fn new_from_wins_losses(wins: u16, losses: u16) -> Self {
+    pub fn new_from_wins_losses(wins: u16, losses: u16) -> Self {
         Self {
-            state: None,
             wins,
             losses,
-            value: 0,
+            ..Default::default()
         }
     }
 
     #[cfg(test)]
-    pub const fn new_from_value(value: i32) -> Self {
+    pub fn new_from_value(value: i32) -> Self {
         Self {
-            state: None,
-            wins: 0,
-            losses: 0,
             value,
+            ..Default::default()
         }
     }
 
     #[cfg(test)]
-    pub const fn new_from_state(state: Option<MVSGameState>) -> Self {
+    pub fn new_from_state(state: Option<MVSGameState>) -> Self {
         Self {
             state,
-            wins: 0,
+            ..Default::default()
-            losses: 0,
-            value: 0,
         }
     }
 
     fn chance_win(&self) -> Option<f32> {
-        let sum = self.losses + self.wins;
+        let sum = self.losses + self.wins + self.ties;
-        if sum == 0 {
+        if 20 > sum {
             return None;
         }
         Some(self.wins as f32 / sum as f32)
@@ -70,12 +66,11 @@ impl MoveValueStats {
                 .iter()
                 .filter(|x| x.state == Some(MVSGameState::Loss))
                 .count() as u16;
-    }
+        self.ties = others.iter().map(|x| x.ties).sum::<u16>()
-}
+            + others
-
+                .iter()
-impl PartialOrd for MoveValueStats {
+                .filter(|x| x.state == Some(MVSGameState::Tie))
-    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+                .count() as u16;
-        Some(self.cmp(other))
     }
 }
 
@@ -122,4 +117,11 @@ mod tests {
         let two = MoveValueStats::new_from_state(Some(MVSGameState::Win));
         assert!(one < two);
     }
+
+    #[test]
+    fn two_prob_zero() {
+        let one = MoveValueStats::new_from_wins_losses(10, 0);
+        let two = MoveValueStats::new_from_wins_losses(0, 6);
+        assert!(one > two);
+    }
 }