Edited model for certification
Browse files- configuration.yaml +32 -0
configuration.yaml
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
behaviors:
|
2 |
+
SoccerTwos:
|
3 |
+
trainer_type: poca
|
4 |
+
hyperparameters:
|
5 |
+
batch_size: 4096 # تغییر از 2048 به 4096
|
6 |
+
buffer_size: 20480
|
7 |
+
learning_rate: 0.0005 # تغییر از 0.0003 به 0.0005
|
8 |
+
beta: 0.005
|
9 |
+
epsilon: 0.2
|
10 |
+
lambd: 0.95
|
11 |
+
num_epoch: 3
|
12 |
+
learning_rate_schedule: constant
|
13 |
+
network_settings:
|
14 |
+
normalize: false
|
15 |
+
hidden_units: 512
|
16 |
+
num_layers: 2
|
17 |
+
vis_encode_type: simple
|
18 |
+
reward_signals:
|
19 |
+
extrinsic:
|
20 |
+
gamma: 0.99
|
21 |
+
strength: 1.0
|
22 |
+
keep_checkpoints: 5
|
23 |
+
max_steps: 5000000
|
24 |
+
time_horizon: 1000
|
25 |
+
summary_freq: 10000
|
26 |
+
self_play:
|
27 |
+
save_steps: 50000
|
28 |
+
team_change: 200000
|
29 |
+
swap_steps: 2000
|
30 |
+
window: 10
|
31 |
+
play_against_latest_model_ratio: 0.5
|
32 |
+
initial_elo: 1200.0
|