| |
| reward_specs: |
| dis_weight: 0.5 |
| cost_weight: 0.5 |
| build_ration: 0.5 |
|
|
| |
| agent_specs: |
| batch_stage: false |
|
|
| gamma: 0.9 |
| tau: 0.0 |
| state_encoder_specs: |
| state_encoder_hidden_size: [16, 4] |
| gcn_node_dim: 16 |
| num_gcn_layers: 2 |
| num_edge_fc_layers: 1 |
| max_num_nodes: 1000 |
| max_num_edges: 3000 |
| num_attention_heads: 1 |
| policy_specs: |
| policy_land_use_head_hidden_size: [32, 1] |
| policy_road_head_hidden_size: [32, 1] |
| value_specs: |
| value_head_hidden_size: [32, 32, 1] |
| lr: 4.0e-4 |
| weightdecay: 0.0 |
| eps: 1.0e-5 |
|
|
| value_pred_coef: 0.5 |
| entropy_coef: 0.01 |
| clip_epsilon: 0.2 |
|
|
|
|
| max_num_iterations: 100 |
| |
| num_episodes_per_iteration: 1200 |
| |
| max_sequence_length: 33 |
| num_optim_epoch: 4 |
| |
| mini_batch_size: 1024 |
| save_model_interval: 1 |
|
|
