🐐

backend/train_nn.py

@@ -20,8 +20,6 @@ P2 = "p2"
 FIXED_PERSONALITIES = [p for p in AIPersonality if p != AIPersonality.ARBITRARY]
 
 
-# ==================== Game runner ====================
-
 def _build_player(pid: str, name: str, cards: list, difficulty: int, personality: AIPersonality) -> PlayerState:
   deck = choose_cards(cards, difficulty, personality)
   instances = _make_instances(deck)
@@ -82,8 +80,6 @@ def run_episode(
   return state.result.winner_id if state.result else None
 
 
-# ==================== Training loop ====================
-
 def train(
   n_episodes: int = 20_000,
   self_play_start: int = 5_000,
@@ -124,7 +120,6 @@ def train(
     nn_goes_first = random.random() < 0.5
 
     if random.random() < self_play_prob:
-      # ---- Self-play ----
       nn1 = NeuralPlayer(net, training=True, temperature=temperature)
       nn2 = NeuralPlayer(net, training=True, temperature=temperature)
 
@@ -148,7 +143,6 @@ def train(
           batch_count += 1
 
     else:
-      # ---- NN vs fixed opponent ----
       opp_personality = random.choice(FIXED_PERSONALITIES)
       nn_player = NeuralPlayer(net, training=True, temperature=temperature)
       opp_ctrl  = lambda p, o, pers=opp_personality, diff=opp_difficulty: choose_plan(p, o, pers, diff)