/FrFTimeSeries

The repository of the paper Fractional Fourier Transform in Time Series submitted to IEEE Signal Processing Letters.

Primary LanguagePython

Fractional Fourier Transform in Time Series Prediction

This is the GitHub repository for the paper: E.Koc, A. Koç, “ Fractional Fourier Transform in Time Series Prediction ” accepted to IEEE Signal Processing Letters, 2022. In this study, we apply a window function to a univariate time series and divide it into segment. Then, we apply fractional Fourier transform on each segment and obtain feature vectors. These feature vectors are fed into GRU/RNN based encoder-decoder to predict the future time series. This repository is benefited from the codes in the paper Sequence Prediction using Spectral RNNs and modified the codes depending on the tasks. Codes are implemented using Tensorflow 1.15 (1.1x is also suitable )on Linux based systems. You can read our paper on Fractional Fourier Transform in Time Series Prediction.

Training

Run the script python3 frft_test.py to train the network. There are several important points to modify in frft_test.py for future studies. pd['base_dir'] is the directory that you save your models. You can run the list of multiple experiments and save the models here.

Test

To test the models, run the script python3 mse_test.py.

References

M. Wolter, J. Gall, and A. Yao, “Sequence prediction using spectral RNNs,” in International Conference on Artificial Neural Networks. Springer, 2020, pp. 825–837.