"# NE-BJ500-for-BasicTS" 将NE-BJ数据集用于交通流MTS预测，基于Traffic-Benchmark中公开数据，以BasicTS格式呈现。

1. Download https://github.com/zezhishao/BasicTS and https://github.com/tsinghua-fib-lab/Traffic-Benchmark/tree/master/methods/GMAN/BJ500/data
2. copy BJ500.h5(unzip BJ500.7z) to datasets\raw_data\BJ500\BJ500.h5
3. copy Adj(BJ500).txt to datasets\raw_data\BJ500\Adj(BJ500).txt
4. copy generate_training_data.py to scripts\data_preparation\BJ500\generate_training_data.py
5. copy DCRNN_BJ500.py to examples/DCRNN/DCRNN_BJ500.py
6. run python scripts\data_preparation\BJ500\generate_training_data.py
7. check datasets\BJ500\adj_mx.pkl
8. run model run.py on examples/DCRNN/DCRNN_BJ500.py

Result:

data_file_path = datasets/raw_data/BJ500/BJ500.h5
dom = True
dow = True
graph_file_path = datasets/raw_data/BJ500/Adj(BJ500).txt
history_seq_len = 12
norm_each_channel = False
output_dir = datasets/BJ500
target_channel = [0]
tod = True
train_ratio = 0.7
valid_ratio = 0.1
raw time series shape: (6624, 500, 1) number of training samples:4621 number of validation samples:660 number of test samples:1320 mean (training data): 52.12267299942433 std (training data): 20.76980173577049

test_time: 15.08 (s), test_MAE: 4.5578, test_RMSE: 8.5757, test_MAPE: 0.1573, test_WAPE: 0.0917, test_MSE: 73.5423