/high-order-weighted-DMD

Primary LanguageJupyter NotebookMIT LicenseMIT

High-order weighted dynamic mode decomposition (HW-DMD)

Code for paper "Real-time forecasting of metro origin-destination matrices with high-order weighted dynamic mode decomposition", https://arxiv.org/abs/2101.00466

HW-DMD is a simple yet powerful model for forecasting in high-dimensional time series. The repository includes a small example to facilitate the rapid adoption of HW-DMD to general forecast problems.

How to use

The best way to use HW-DMD in your forecast problem is to modify from the example notebook, where we used HW-DMD to forecast the traffic speed at 323 loop detectors in Seattle. Download the repository and replace the Seattle data with your data, then play with the code.

The Experiments/ folder contains the code used in the HW-DMD paper. Some explanations for the organization:

  • Experiments of the paper are shown in the .ipynb files in the Experiments/ folder.
  • The key module of HW-DMD is the HWDMD class located in Experiments/functions.py.
  • The Hangzhou metro dataset (from https://zenodo.org/record/3145404#.YVUaHTHMKPo) is included in the data/ directory. Experiments on the Hangzhou dataset are runnable from .ipynb files.
  • The Guangzhou metro dataset is not included. Files for experiments are included for illustration purposes.

Environment for reproducible results

  • Python 3.8.8
  • Numpy 1.20.1
  • Scipy 1.6.1

For benchmark models that use tensorflow

  • Tensorflow 2.6.0