victorqueiroz commited on
Commit
53ee278
·
verified ·
1 Parent(s): 082ec94

teacher/v6distill-v1: v6 self-play-imitation distill (UNDERPERFORMED, ladder 1599 ~v5)

Browse files
teacher/v6distill-v1/keras_model.keras ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:372c62aacd231aaeec045ce9423d560b7a2addddcab011d00422734c218ed1ac
3
+ size 295790767
teacher/v6distill-v1/metrics.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "val_metrics": {
3
+ "policy_top1": 0.44727495407225965,
4
+ "policy_top3": 0.7080220453153705,
5
+ "policy_top5": 0.8075015309246785,
6
+ "n": 32660,
7
+ "value_mae": 0.27902138233184814,
8
+ "value_sign_agree": 0.6192896509491733
9
+ },
10
+ "best_val_loss": 2.974766254425049,
11
+ "epochs_ran": 14,
12
+ "train_n": 293947,
13
+ "val_n": 32660,
14
+ "value_scale": 400.0,
15
+ "soft_temp_cp": 150.0,
16
+ "arch": {
17
+ "filters": 256,
18
+ "blocks": 20,
19
+ "policy_conv_filters": 48,
20
+ "value_conv_filters": 8,
21
+ "value_hidden": 384,
22
+ "value_hidden2": 192,
23
+ "policy_loss_weight": 1.0,
24
+ "value_loss_weight": 2.0,
25
+ "value_head": "tanh"
26
+ },
27
+ "params": 24608635,
28
+ "data_mode": "selfplay_pi_z",
29
+ "init_from": "/root/gen1.keras",
30
+ "resumed_from": null
31
+ }
teacher/v6distill-v1/v6distill_ladder.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_a": "/root/v6distill/keras_model.keras",
3
+ "opponent": "stockfish-elo-ladder[1500,1700,1900,2100]",
4
+ "mode": "stockfish_elo_ladder",
5
+ "sims": 160,
6
+ "mcts_batch": 16,
7
+ "sf_nodes": 100000,
8
+ "max_moves": 200,
9
+ "openings": 63,
10
+ "openings_source": "/root/openings_balanced.txt",
11
+ "openings_used": 50,
12
+ "pairs_per_rung": 50,
13
+ "absolute_elo": 1598.8,
14
+ "absolute_elo_ci95": [
15
+ 1452.6,
16
+ 1745.0
17
+ ],
18
+ "absolute_elo_ci95_fixed": [
19
+ 1564.0,
20
+ 1633.6
21
+ ],
22
+ "rung_spread_elo": 333.7,
23
+ "heterogeneity_tau": 144.5,
24
+ "heterogeneity_q": 51.6,
25
+ "n_rungs_used": 4,
26
+ "rungs": [
27
+ {
28
+ "sf_elo": 1500,
29
+ "score_rate": 0.44,
30
+ "net_elo": 1458.1,
31
+ "net_elo_ci95": [
32
+ 1391.4,
33
+ 1521.8
34
+ ]
35
+ },
36
+ {
37
+ "sf_elo": 1700,
38
+ "score_rate": 0.28,
39
+ "net_elo": 1535.9,
40
+ "net_elo_ci95": [
41
+ 1468.9,
42
+ 1593.1
43
+ ]
44
+ },
45
+ {
46
+ "sf_elo": 1900,
47
+ "score_rate": 0.265,
48
+ "net_elo": 1722.8,
49
+ "net_elo_ci95": [
50
+ 1647.1,
51
+ 1785.5
52
+ ]
53
+ },
54
+ {
55
+ "sf_elo": 2100,
56
+ "score_rate": 0.145,
57
+ "net_elo": 1791.8,
58
+ "net_elo_ci95": [
59
+ 1684.6,
60
+ 1865.8
61
+ ]
62
+ }
63
+ ]
64
+ }