TrafficWheel/config/AGCRN/AirQuality.yaml

55 lines
793 B
YAML

basic:
dataset: AirQuality
device: cuda:0
mode: train
model: AGCRN
seed: 2023
data:
batch_size: 16
column_wise: false
days_per_week: 7
horizon: 24
input_dim: 6
lag: 24
normalizer: std
num_nodes: 12
steps_per_day: 24
test_ratio: 0.2
val_ratio: 0.2
model:
cheb_k: 2
cheb_order: 2
embed_dim: 10
input_dim: 6
num_layers: 2
num_nodes: 12
output_dim: 6
rnn_units: 64
train:
batch_size: 16
debug: false
early_stop: true
early_stop_patience: 15
epochs: 100
grad_norm: false
log_step: 100
loss_func: mae
lr_decay: false
lr_decay_rate: 0.3
lr_decay_step:
- 5
- 20
- 40
- 70
lr_init: 0.003
mae_thresh: 0.0
mape_thresh: 0.0
max_grad_norm: 5
output_dim: 6
plot: false
real_value: true
weight_decay: 0