huawei-noah / BHT-ARIMA

Code for paper: Block Hankel Tensor ARIMA for Multiple Short Time Series Forecasting (AAAI-20)

BHT-ARIMA

A tensor decomposition-based time series forecasting algorithm, which tactically incorporates the unique advantages of Hankelization, low-rank Tucker decomposition and ARIMA into a unified framework.
More details (including parameter settings) refer to the original paper.

Paper

• "Block Hankel Tensor ARIMA for Multiple Short Time Series Forecasting", AAAI-20

Datasets

Traffic dataset. The traffic data is originally collected from California department of transportation 1 and describes the road occupy rate of Los Angeles County highway network.We here use the same subset used in (Yu, Yin, and Zhu 2017) which selects 228 sensors randomly. And We take the first 40 time points of them as data of our demo

Getting Started

Prerequisites

• python 3.5+
• python libraries
  • tensorly
  • scipy
  • numpy
  • pandas

Run

python main.py

License

© Contributors, 2019. Licensed under an MIT license.