Deep CFR 무제한 학습 config 추가

configs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_unbounded.yaml

@@ -0,0 +1,94 @@
+run:
+  experiment_name: lost_cities_deep_cfr_selfplay_full_depth_slot_playability_unbounded
+  iterations: null
+  seed: 79
+  max_iterations: null
+  max_hours: null
+  device: cuda
+  use_amp: false
+
+rules:
+  n_colors: 5
+  n_ranks: 9
+  min_rank: 2
+  n_handshakes: 3
+  hand_size: 8
+  expedition_penalty: -20
+  bonus_threshold: 8
+  bonus_amount: 20
+
+encoding:
+  derived_playability: true
+  slot_aware_playability: true
+
+network:
+  hidden_size: 256
+  num_layers: 3
+  activation: relu
+
+traversal:
+  traversals_per_player: 70
+  strategy_sample_interval: 1
+  store_strategy_on_opponent_nodes: false
+  store_strategy_on_traverser_nodes: true
+  max_depth: null
+  max_nodes_per_traversal: 1000
+  opponent_policy: self_play_league
+  cutoff_value_mode: score_diff
+  cutoff_rollouts: 0
+  cutoff_rollout_policy: random
+  cutoff_rollout_max_steps: 300
+  progress_every_traversals: 10
+  num_workers: 8
+  traversal_worker_chunk_size: 8
+  regret_matching_epsilon: 0.0001
+  outcome_sampling_epsilon: 0.2
+  outcome_sampling_value_clip: 500
+  outcome_unsampled_regret: zero
+  endpoint_depth_bucket_width: 100
+  endpoint_depth_bucket_max: 1000
+
+self_play:
+  current_weight: 0.5
+  recent_weight: 0.3
+  older_weight: 0.2
+  anchor_weight: 0.0
+  recent_window: 5
+  max_snapshots: 20
+  snapshot_every: 1
+
+optimization:
+  advantage_batch_size: 1024
+  strategy_batch_size: 1024
+  advantage_updates_per_iteration: 256
+  strategy_updates_per_iteration: 256
+  learning_rate: 0.00003
+  weight_decay: 0.0001
+  grad_clip: 1.0
+
+memory:
+  advantage_capacity: 2000000
+  strategy_capacity: 2000000
+
+evaluation:
+  eval_every: 5
+  games: 100
+  batch_size: 64
+  device: trainer
+  num_workers: 4
+  max_steps: 1000
+  on_max_steps: score_diff
+  opponents:
+    - random
+    - passive_discard
+    - safe_heuristic
+    - safe_heuristic_loose
+    - safe_heuristic_strict
+    - noisy_safe
+
+checkpoint:
+  directory: runs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_unbounded
+  save_every_iteration: false
+  save_iteration_interval: 100
+  save_latest_only: false
+  progress_interval_seconds: 20.0

src/coolrl_lost_cities/games/classic/deep_cfr/config.py

@@ -18,7 +18,7 @@ class StrictModel(BaseModel):
 
 class RunConfig(StrictModel):
     experiment_name: str = "deep_cfr"
-    iterations: int = 1
+    iterations: int | None = 1
     max_iterations: int | None = None
     max_hours: float | None = None
     seed: int = 1

src/coolrl_lost_cities/games/classic/deep_cfr/trainer.py

@@ -534,6 +534,8 @@ class DeepCFRTrainer:
             return max(self.iteration, int(self.config.run.max_iterations))
         if self.config.run.max_hours is not None:
             return 2**31 - 1
+        if self.config.run.iterations is None:
+            return 2**31 - 1
         return self.iteration + self.config.run.iterations
 
     def _time_limit_reached(self, run_started: float) -> bool: