fly-xox-malecns / training /results.json
n4ze3m's picture
Preserve pilot checkpoints and results
f5e3dc9 verified
Raw
History Blame
1.53 kB
{
"validation_selected_step": 1100,
"elapsed_seconds": 42.79798544710502,
"training": {
"positions": 3560,
"optimal_move_rate": 0.9817415475845337,
"illegal_moves": 0
},
"validation": {
"positions": 492,
"optimal_move_rate": 0.9004064798355103,
"illegal_moves": 0
},
"test": {
"positions": 468,
"optimal_move_rate": 0.9166666865348816,
"illegal_moves": 0
},
"trainable_parameters": 500100,
"changed_internal_parameters": {
"gain": 153090,
"bias": 165455,
"leak": 165455
},
"fixed_interfaces_unchanged": true,
"one_training_seed": true,
"split": "D4 symmetry groups of player-relative boards, SHA256 bucket; 80/10/10 approximate",
"device": "NVIDIA GeForce RTX 3090",
"peak_gpu_gb": 1.013288448,
"uniform_random_optimal_rate": 0.5691086691086691,
"untrained_test": {
"positions": 468,
"optimal_move_rate": 0.6282051205635071,
"illegal_moves": 0
},
"games": {
"random_fly_first": {
"wins": 474,
"draws": 26,
"losses": 0
},
"random_fly_second": {
"wins": 425,
"draws": 70,
"losses": 5
},
"minimax_fly_first": {
"wins": 0,
"draws": 500,
"losses": 0
},
"minimax_fly_second": {
"wins": 0,
"draws": 480,
"losses": 20
}
},
"games_note": "500 games per opponent and side, deterministic learned policy, seeded random opponent/tie choices. Complete games can include training boards; held-out position accuracy is reported separately."
}