diff --git a/cfg/algo/ppo.yaml b/cfg/algo/ppo.yaml
index a23c1e3..cdd3f1c 100644
--- a/cfg/algo/ppo.yaml
+++ b/cfg/algo/ppo.yaml
@@ -20,7 +20,7 @@ optimizer:
     lr: 5e-4
 
 
-num_channels: 32
-num_residual_blocks: 3
+num_channels: 64
+num_residual_blocks: 4
 
 
diff --git a/cfg/baseline/ppo.yaml b/cfg/baseline/ppo.yaml
index 541fcaf..7ac6ec8 100644
--- a/cfg/baseline/ppo.yaml
+++ b/cfg/baseline/ppo.yaml
@@ -20,7 +20,7 @@ optimizer:
     lr: 3e-4
 
 
-num_channels: 32
-num_residual_blocks: 3
+num_channels: 64
+num_residual_blocks: 4
 
 
diff --git a/cfg/train_InRL.yaml b/cfg/train_InRL.yaml
index 43a30e6..30bc74e 100644
--- a/cfg/train_InRL.yaml
+++ b/cfg/train_InRL.yaml
@@ -7,13 +7,13 @@ run_dir:
 
 augment: false
 
-epochs: 1500
+epochs: 1000
 rounds: 64
-save_interval: -1
+save_interval: 300
 
 
-black_checkpoint: pretrained_models/${board_size}_${board_size}/${algo.name}/0.pt
-white_checkpoint: pretrained_models/${board_size}_${board_size}/${algo.name}/1.pt
+black_checkpoint: black_final.pt # pretrained_models/${board_size}_${board_size}/${algo.name}/0.pt
+white_checkpoint: white_final.pt # pretrained_models/${board_size}_${board_size}/${algo.name}/1.pt
 
 wandb:
   group: ${board_size}_${board_size}_${algo.name}_InRL
diff --git a/gomoku_rl/utils/elo.py b/gomoku_rl/utils/elo.py
index f653225..7207d45 100644
--- a/gomoku_rl/utils/elo.py
+++ b/gomoku_rl/utils/elo.py
@@ -19,3 +19,27 @@ def compute_elo_ratings(payoff: np.ndarray, average_rating: float = 1200) -> np.
     elo_ratings = payoff.mean(axis=-1)
     elo_ratings = elo_ratings * (400 / np.log(10)) + average_rating
     return elo_ratings
+
+
+def compute_expected_score(rating_0: float, rating_1: float) -> float:
+    return 1 / (1 + 10.0 ** ((rating_1 - rating_0) / 400))
+
+
+class Elo:
+    def __init__(self) -> None:
+        self.players: dict[str, float] = {}
+
+    def addPlayer(self, name: str, rating: float = 1200):
+        assert name not in self.players
+        self.players[name] = rating
+
+    def expected_score(self, player_0: str, player_1: str) -> float:
+        rating_0 = self.players[player_0]
+        rating_1 = self.players[player_1]
+        return compute_expected_score(rating_0, rating_1)
+
+    def update(self, player_0: str, player_1: str, score: float, K: float = 64):
+        e = self.expected_score(player_0, player_1)
+        tmp = K * (score - e)
+        self.players[player_0] = self.players[player_0] + tmp
+        self.players[player_1] = self.players[player_1] - tmp
diff --git a/pretrained_models/15_15/ppo/0.pt b/pretrained_models/15_15/ppo/0.pt
index da4fd9a..0e3b99f 100644
Binary files a/pretrained_models/15_15/ppo/0.pt and b/pretrained_models/15_15/ppo/0.pt differ
diff --git a/pretrained_models/15_15/ppo/1.pt b/pretrained_models/15_15/ppo/1.pt
index c5c64b2..e2c96f6 100644
Binary files a/pretrained_models/15_15/ppo/1.pt and b/pretrained_models/15_15/ppo/1.pt differ