Clean up Deep CFR config schema

Remove legacy aliases, rename max_hours to max_minutes, collapse the
four checkpoint save flags into save_every + save_latest, and change
defaults to safer values (opponent_policy=self_play_league,
device=auto, eval_every=50, max_depth=null). Migrate all archived
yaml configs and tests to the new schema.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

configs/deep_cfr/deep_cfr_color_shared_512x3.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_color_shared_512x3
-  iterations: null
   seed: 42
   max_iterations: 10000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -30,10 +29,10 @@ network:
   color_attention_heads: 4
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -49,7 +48,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -71,9 +70,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -89,9 +87,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_color_shared_512x3
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_color_shared_attention_512x3.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_color_shared_attention_512x3
-  iterations: null
   seed: 42
   max_iterations: 10000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -30,10 +29,10 @@ network:
   color_attention_heads: 8
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -49,7 +48,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -71,9 +70,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -89,9 +87,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_color_shared_attention_512x3
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_color_shared_attention_exp_1000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_color_shared_attention_1000iter
-  iterations: null
   seed: 42
   max_iterations: 1000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -30,10 +29,10 @@ network:
   color_attention_heads: 8
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -49,7 +48,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -71,9 +70,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 256
@@ -89,9 +87,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_color_shared_attention_exp_1000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_opponent_average_strategy_512x3_1000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_opponent_average_strategy_512x3_1000iter
-  iterations: null
   seed: 79
   max_iterations: 1000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_opponent_average_strategy_512x3_1000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_opponent_network_1024x4_1000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_opponent_network_1024x4_1000iter
-  iterations: null
   seed: 79
   max_iterations: 1000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_opponent_network_1024x4_1000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_opponent_network_512x3_1000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_opponent_network_512x3_1000iter
-  iterations: null
   seed: 79
   max_iterations: 1000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_opponent_network_512x3_1000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_pure_selfplay_full_depth_slot_playability_512x3_2x_updates_10000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_pure_selfplay_full_depth_slot_playability_512x3_2x_updates_10000iter
-  iterations: null
   seed: 79
   max_iterations: 10000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_pure_selfplay_full_depth_slot_playability_512x3_2x_updates_10000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_selfplay_anchor_safe_512x3_2x_updates_10000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_selfplay_anchor_safe_512x3_2x_updates_10000iter
-  iterations: null
   seed: 79
   max_iterations: 10000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_selfplay_anchor_safe_512x3_2x_updates_10000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability.yaml

@@ -2,7 +2,7 @@ run:
   experiment_name: lost_cities_deep_cfr_selfplay_full_depth_slot_playability
   seed: 79
   max_iterations: null
-  max_hours: 4
+  max_minutes: 240
   device: cuda
   use_amp: false
 
@@ -39,7 +39,7 @@ traversal:
   cutoff_rollout_max_steps: 300
   progress_every_traversals: 10
   num_workers: 8
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
   outcome_sampling_value_clip: 500
@@ -87,7 +87,5 @@ evaluation:
 
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability
-  save_every_iteration: false
-  save_iteration_interval: 10
-  save_latest_only: false
+  save_every: 10
   progress_interval_seconds: 20.0

configs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_512x3_2x_updates_10000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_selfplay_full_depth_slot_playability_512x3_2x_updates_10000iter
-  iterations: null
   seed: 79
   max_iterations: 10000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 512
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_512x3_2x_updates_10000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_512x3_lcfr_10000iter.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_selfplay_full_depth_slot_playability_512x3_lcfr_10000iter
-  iterations: null
   seed: 79
   max_iterations: 10000
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -69,9 +68,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 256
@@ -87,9 +85,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_512x3_lcfr_10000iter
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_512x3_unbounded.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_selfplay_full_depth_slot_playability_512x3_unbounded
-  iterations: null
   seed: 79
   max_iterations: null
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -27,10 +26,10 @@ network:
   activation: relu
 
 traversal:
-  traversals_per_iteration: 2
+  traversals_per_player: 2
   traversals_per_player: 70
   max_depth: null
-  max_nodes: 10000
+  max_nodes_per_traversal: 10000
   max_nodes_per_traversal: 1000
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
@@ -46,7 +45,7 @@ traversal:
   store_strategy_on_opponent_nodes: false
   num_workers: 8
   worker_chunk_size: 4
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   progress_every_traversals: 10
   endpoint_depth_bucket_width: 100
   endpoint_depth_bucket_max: 1000
@@ -68,9 +67,8 @@ self_play:
   recent_window: 5
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 32
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   advantage_batch_size: 1024
   strategy_batch_size: 1024
   advantage_updates_per_iteration: 256
@@ -86,9 +84,7 @@ memory:
 checkpoint:
   directory: runs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_512x3_unbounded
   save_latest: true
-  save_every_iteration: false
-  save_iteration_interval: 100
-  save_latest_only: false
+  save_every: 100
   progress_interval_seconds: 20.0
   exact_resume: false
 

configs/deep_cfr/deep_cfr_selfplay_full_depth_slot_playability_unbounded.yaml

@@ -1,9 +1,8 @@
 run:
   experiment_name: lost_cities_deep_cfr_selfplay_full_depth_slot_playability_unbounded
-  iterations: null
   seed: 79
   max_iterations: null
-  max_hours: null
+  max_minutes: null
   device: cuda
   use_amp: false
 
@@ -40,7 +39,7 @@ traversal:
   cutoff_rollout_max_steps: 300
   progress_every_traversals: 10
   num_workers: 8
-  traversal_worker_chunk_size: 8
+  worker_chunk_size: 8
   regret_matching_epsilon: 0.0001
   outcome_sampling_epsilon: 0.2
   outcome_sampling_value_clip: 500
@@ -88,7 +87,5 @@ evaluation:
 
 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
+  save_every: 100
   progress_interval_seconds: 20.0

configs/deep_cfr/smoke.yaml

@@ -1,5 +1,5 @@
 run:
-  iterations: 1
+  max_iterations: 1
   seed: 1
   device: cpu
 
@@ -7,16 +7,17 @@ network:
   hidden_size: 16
 
 traversal:
-  traversals_per_iteration: 1
+  traversals_per_player: 1
   max_depth: 2
-  max_nodes: 64
+  max_nodes_per_traversal: 64
   num_workers: 0
   worker_chunk_size: 1
 
 optimization:
-  advantage_train_steps: 1
-  strategy_train_steps: 1
-  batch_size: 2
+  advantage_batch_size: 2
+  strategy_batch_size: 2
+  advantage_updates_per_iteration: 1
+  strategy_updates_per_iteration: 1
   learning_rate: 0.001
 
 memory:
@@ -25,7 +26,8 @@ memory:
 
 checkpoint:
   directory: runs/deep_cfr/smoke
-  save_every_iteration: false
+  save_every: 0
+  save_latest: false
 
 evaluation:
   eval_every: 0