Table 1.
Deep reinforcement learning hyperparameters.
| Parameters | Deep Reinforcement Learning’s Value |
|---|---|
| A | 0.50 |
| Batch size | 20 |
| Block size | 1000 |
| Epochs | 500 |
| Generation epoch | 1040 |
| GGNN activation function | SELU |
| GGNN depth | 4 |
| GGNN dropout probability | 0 |
| GGNN hidden dimension | 250 |
| GGNN width | 100 |
| Initial learning rate | 1.00 × 10−4 |
| Learning rate decay factor | 0.99 |
| Learning rate decay interval | 10 |
| Loss function | Kullback–Leibler divergence |
| Maximum relative learning rate | 1.00 |
| Message passing layers | 3 |
| Message size–input size of GRU | 100 |
| Minimum relative learning rate | 1.00 × 10−4 |
| MLP activation function | SoftMax |
| MLP depth(Layers 1 and 2) | 4 |
| MLP dropout probability (Layers 1 and 2) |
0 |
| MLP hidden dimension (Layers 1 and 2) |
500 |
| Number of samples | 200 |
| Optimizer | Adam |
| 20 | |
| Weight decay | 0 |
| Weight initialization | Uniform |