guoshnbjtu/sstban

SSTBAN model

We give three examples (PEMSD8_36_36, PEMSD4_36_36, Seattle_1hour_36_36), and hyperparameters in the configuration documents are the best.

• PEMSD8 experiments:

  • data prepare:

  • 1. Enter the "SSTBAN/prepare/"
  • 2. Run "python prepareData_SSTBAN.py --config configurations/PEMSD8_1dim_construct_samples.conf"
  • 3. We can get different training data by adjusting the PEMSD8_1dim_construct_samples.conf

  • model training:

  • 1. Enter the "SSTBAN/"
  • 2. Run "python train_SSTBAN.py --config configurations/PEMSD8_1dim_24.conf"

• PEMSD4 experiments:

  • data prepare:

  • 1. Enter the "SSTBAN/prepare/"
  • 2. Run "python prepareData_SSTBAN.py --config configurations/PEMSD4_1dim_construct_samples.conf"
  • 3. We can get different training data by adjusting the PEMSD4_1dim_construct_samples.conf

  • model training:

  • 1. Enter the "SSTBAN/"
  • 2. Run "python train_SSTBAN.py --config configurations/PEMSD4_1dim_24.conf"

• Seattle experiments:

  • data prepare:

  • 1. Enter the "SSTBAN/prepare/"
  • 2. Run "python prepareData_SSTBAN.py --config configurations/Seattle_3dim_construct_samples.conf"
  • 3. We can get different training data by adjusting the Seattle_3dim_construct_samples.conf

  • model training:

  • 1. Enter the "SSTBAN/"
  • 2. Run "python train_SSTBAN.py --config configurations/Seattle_3dim_24.conf"

You can download the data from:

Google Drive: https://drive.google.com/file/d/1vBn8UDbIOz5T_MAahVNoYccR0jDjjFh-/view?usp=sharing

One Drive: https://1drv.ms/u/s!AuEMUxDDxoXMxGdrVE7f9CYy330R?e=l6Jy01

Baidu Cloud: https://pan.baidu.com/s/1MwWspHYSsg5fJsiYMiqn2g password：g4hw

Aliyun Cloud: https://www.aliyundrive.com/s/1bkb3bUFE53

sstban's Issues

关于数据集数据的含义

您好！感谢您开源模型和论文，我下载了Seattle数据集，加载了npz文件查看，您是否方便介绍一下这个三维数组的每个维度代表什么含义，我在论文中和代码中都没有找到相关内容

shape: (8760, 323, 5)
data:
[65.45061086727755 0.028903472222222226 613.0 0
 Timestamp('2015-01-01 00:00:00')]

Inconsistency in the settings

Hi, Shengnan thanks for sharing the code and hope you are doing well. Just need some clarification on the setting of two hyper-parameters, 1) the number of attention heads and 2) the dimension of each attention head outputs. It seems that they are not consistent between the code and the paper (e.g., for PEMSD4_24, in the config file, K=16, d=8; while in TABLE III of the paper, h=8, d=16).

Can you please confirm which configuration is the right one?

Where can I download your paper from ?

I can't found your paper Self-Supervised Spatial-Temporal Bottleneck Attentive Network for Efficient Long-term Traffic Forecasting at IEEE or Google Scholar.