Code for representing the dynamics of spatiotemporal data with non-redundant wavelets

Reference: “Representing the dynamics of high-dimensional data with non-redundant wavelets” by Shanshan Jia, Xingyi Li, Tiejun Huang, Jian K. Liu, Zhaofei Yu

Linux system installed CMICOT and Matlab

To run the WCMI method, one needs to add two following packages (both were cloned here):

1. WItoobox [https://www2.le.ac.uk/centres/csn/software/WI]: here one can implement the wavelet decomposition and compute the mutual formation of each feature.
2. CMICOT [https://github.com/yandex/CMICOT]: we have updated the code, so that one can output the ranking ID of each feature, together with the conditional mutual information value of each feature.

1. Run WCMI_caller.m to save all the wavelet coefficients in a txt file. You can change to other data of interest inside.

   One can also run WCMI_population_caller.m with Experimental data - 'Retinal ganglion cell responses to natural images'.

2. Run ' ./cmicot --pool data > data_feature_ranking_6s --select-count k' in terminal, to save the top k coefficients selected by CMI in a new txt. Change data to your data file of interest.

3. Run WCMI_decoding.m to get the decoding results of the data of interest.

1. Simulated data: The data folder contains the simulated data taken from WItoobox.

2. Experimental data - 'Retinal ganglion cell responses to natural images' can be found at [https://doi.org/10.12751/g-node.kod28e].

3. Experimental data - 'Human ECoG speaking consonant-vowel syllables' can be found at [https://doi.org/10.6084/m9.figshare.c.4617263.v4].