Home

This is a support page of the papers Zhiguang Wang and Tim Oates. 2015. Imaging time-series to improve classification and imputation. In Proceedings of the 24th International Conference on Artificial Intelligence (IJCAI'15), Qiang Yang and Michael Wooldridge (Eds.). AAAI Press 3939-3945.

Wang, Zhiguang, and Tim Oates. "Encoding Time Series as Images for Visual Inspection and Classification Using Tiled Convolutional Neural Networks." In Workshops at the Twenty-Ninth AAAI Conference on Artificial Intelligence. 2015.

The scripts are written in Python with the dependencies:

1 Numpy
2 Pandas
3 Matplolib
4 Cpickle

Anaconda is highly recommended since it seals all the dependency quite well.

serie2GAF.py
Encoding time series to GADF/GASF.
The parameters are:

datafiles = ['Coffee_ALL'] # Data fine name
trains = [28] # Number of training instances (because we assume training and test data are mixed in one file)
size = [64] # PAA size
GAF_type = 'GASF' # GAF type: GASF, GADF
save_PAA = True # Save the GAF with or without dimension reduction by PAA: True, False
rescale_type = 'Zero' # Rescale the data into [0,1] or [-1,1]: Zero, Minusone

serie2MTF.py
Encoding time series to MTF.
The parameters are:

Data file name

datafiles = ['Coffee_ALL']

Number of training instances (because we assume training and test data are mixed in one file)

trains = [28]

PAA size

size = [64]

Quantile size

quantile = [16]

Reduce the image size using: full, patch, paa

reduction_type = 'patch'

The output will be an example grach (of the kth instance) of the polar coordnate images, generated GASF/GADF/MTF images and the pkl file.

If you find the script is helpful, you are highly appreciated if you can cite the above two papers.

Any questions please contact Zhiguang Wang by [email protected]. Thanks!