Encoding Time Series in Gram Matrix Formats for Deep Learning

Deep Learning is a very robust family of algorithms for modeling visual information. Several published studies show the advantages of encoding non-visual data into matrix structures. Those methods might offer benefits over 1D-CNNs. This paper is an excellent overview of building Gram Matrix data representation from time-series input. I agree with the authors that temporal correlations can be captured in a way that uses relative correlation by superposition of directions concerning time intervals. The #numpy implementation and a Gif image generation with #matplotlib is an excellent way to go. 

https://medium.com/analytics-vidhya/encoding-time-series-as-images-b043becbdbf3

#1DCNN #gram #grammatrix #gramianangularfield #angularfield #numpy