examples/AutoPower_IEEE39_bus/AdaptiveHierarchicalGraphTransformer/configs/test_senseflow_39.yaml

log_path: ./results
flag_return_losses: True

pq_len: &pq_len 29
pv_len: &pv_len 9
slack_len: &slack_len 1
mask_num: &mask_num 0
batch_size: &batch_size 256

data:
  meta: 
    node: ['PQ', 'PV', 'Slack']
    edge:
    - ['PQ', 'default', 'PQ']
    - ['PQ', 'default', 'PV']
    - ['PQ', 'default', 'Slack']
    - ['PV', 'default', 'PQ']
    - ['PV', 'default', 'PV']
    - ['PV', 'default', 'Slack']
    - ['Slack', 'default', 'PQ']
    - ['Slack', 'default', 'PV']
  train:
    dataset_type: PowerFlowDataset
    data_root: /
    split_txt: ./datasets/power/case39_data/10w_case39_n_n_1.json
    pq_len: *pq_len
    pv_len: *pv_len
    slack_len: *slack_len
    mask_num: *mask_num
  val:
    dataset_type: PowerFlowDataset
    data_root: /
    split_txt: ./datasets/power/case39_data/2w_case39_n_2.json
    pq_len: *pq_len
    pv_len: *pv_len
    slack_len: *slack_len
    mask_num: *mask_num
  batch_size: *batch_size
  batch_size_test: *batch_size
  num_workers: 4

train:
  logs_freq: 10
  epochs: 100
  optimizer_type: "Adam"
  learning_rate: 0.001
  momentum: 0.9
  weight_decay: 0.0

model:
  type: senseflow
  hidden_channels: 128
  num_block: 4
  layers_per_graph: 2
  heads_ca: 8
  batch_size: *batch_size
  flag_use_edge_feat: False
  with_norm: True
  num_loops_train: 1
  num_loops_test: -1
  scaling_factor_vm: 0.01
  scaling_factor_va: 0.01
  loss_type: l1
  flag_weighted_loss: True
  loss_weight_equ: 0.1
  loss_weight_vm: 10.0
  loss_weight_va: 1.0
  matrix: vm_va
  resume_ckpt_path: ""
  flag_use_ema: True
  ema_warmup_epoch: 10
  ema_decay_param: 0.99


scheduler:
  type: Cosine
  eta_min: 1e-5


loss:
  type: bi_deltapq_loss
  filt_type: True
  aggr: abs