https://github.com/VachelHU/EvoNet

Time-Series Event Prediction with Evolutionary State Graph, WSDM 2021
https://github.com/VachelHU/EvoNet

deep-learning time-series-analysis

Last synced: about 1 month ago
JSON representation

Time-Series Event Prediction with Evolutionary State Graph, WSDM 2021

• Host: GitHub
• URL: https://github.com/VachelHU/EvoNet
• Owner: VachelHU
• Created: 2019-05-10T05:54:45.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2021-01-04T02:49:38.000Z (almost 4 years ago)
• Last Synced: 2024-06-23T05:59:37.320Z (5 months ago)
• Topics: deep-learning, time-series-analysis
• Language: Python
• Homepage:
• Size: 66.4 MB
• Stars: 87
• Watchers: 7
• Forks: 25
• Open Issues: 15
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• StarryDivineSky - VachelHU/EvoNet - Series Event Prediction with Evolutionary State Graph 将时间序列转化为动态图进行表示的方法。该方法成功在阿里云 ·SLS 商业化，作为一项智能巡检服务，可以对大规模时间序列进行异常检测与分析。 (时间序列 / 网络服务_其他)

README

        
# EvoNet

This project implements the Evolutionary State Graph Neural Network proposed in [1], which is a GNN-based method for time-series event prediction.

## Compatibility

Code is compatible with tensorflow version 1.2.0 and Pyhton 3.6.2.

Some Python module dependencies are listed in `requirements.txt`, which can be easily installed with pip:

```

pip install -r requirements.txt

```

### Input Format 

An example data format is given where data is stored as a list containing 4 dimensionals tensors such as

 

`[number of samples × segment number × segment length × dimension of observation]`

### Configuration

We can use `./model_core/config.py` to set the parameters of model.

```

class ModelParam(object):

    # basic

    model_save_path = "./model"

    n_jobs = os.cpu_count()

    # dataset

    data_path = './data'

    data_name = 'webtraffic'

    his_len = 15

    segment_len = 24

    segment_dim = 2

    n_event = 2

    norm = True

    # state recognition

    n_state = 30

    covariance_type = 'diag'

    # model

    graph_dim = 256

    node_dim = 96

    learning_rate = 0.001

    batch_size = 1000

    id_gpu = '0'

    pos_weight = 1.0

```

### Main Script

```

python run.py -h

usage: run.py [-h] [-d {djia30, webtraffic}] [-g GPU]

optional arguments:

  -h, --help            show this help message and exit

  -d {djia30,webtraffic}, --dataset {djia30,webtraffic} select the dataset

  -g GPU, --gpu GPU     target gpu id

```

## Reference

[1] Wenjie, H; Yang, Y; Ziqiang, C; Carl, Y and Xiang, R, 2021, Time-Series Event Prediction with Evolutionary State Graph, In WSDM, 2021

```

@inproceedings{hu2021evonet, 

    title={Time-Series Event Prediction with Evolutionary State Graph},

    author={Wenjie Hu and Yang Yang and Ziqiang Cheng and Carl Yang and Xiang Ren},

    booktitle={Proceedings of WSDM},

    year={2021}

}

```