snudatalab / pft

Fast and Accurate Partial Fourier Transform for Time Series Data (KDD 2021)

PFT

This project is a C++ implementation of Fast and Accurate Partial Fourier Transform for Time Series Data (KDD 2021). The codes for FFTW, MKL, Pruned FFT, and Goertzel algorithm are also included in src/.

Prerequisites

The implementation requires the following libraries.

• mkl.h
• mkl_dfti.h
• ipp.h
• ipps.h
• fftw3.h

Datasets

The four datasets used in our paper are available here. They include synthetic random vectors of length of integer power of two. Urban sound contains 4347 sound recordings in urban environment, and Air condition is composed of 29 time-series vectors of air condition information (e.g., temperature and humidity). Stock is a new public data we release; it consists of the daily historical stock prices of FANG, the four American technology companies Facebook, Amazon, Netflix, and Google. We collected closing prices adjusted for stock splits, from 2017-01-03 to 2021-01-08.

Reference

If you use this code, please cite the following paper.

@inproceedings{ParkJK21,
  author    = {Yong-chan Park and
               Jun-Gi Jang and
               U Kang},
  title     = {Fast and Accurate Partial Fourier Transform for Time Series Data},
  booktitle = {ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD)},
  year      = {2021}
}