Table 2 Parameter settings for simulation experiment 1.
From: Autonomous air combat decision making via graph neural networks and reinforcement learning
Training parameters | Values |
|---|---|
act hidden size | 128 128 |
recurrent hidden layers | 1 |
recurrent hidden size | 128 |
buffer size | 3,000 |
clip param | 0.2 |
entropy coef | 0.001 |
gae lambda | 0.95 |
gamma | 0.99 |
lr | 0.0003 |
max grad norm | 2 |
num mini batch | 5 |
num env steps | 100,000,000 |
