Time series classification using CNNs: the Mcfly library

In this note we will briefly explain how to use Convolutional Neural Networks for time series classification. In a simple way, the convolution operation can be seen as applying a linear filter over the time series.

The problem considered here is the following: imagine that we are given sub-sequences of a periodic signal, let's say a sine function. However, sometimes one sub-sequence is different, as you can see in the figure above: one of the "periods" was replaced by a square signal.

We are going to use the Mcfly library, available here:

introduced in this paper:

D. van Kuppevelt, C. Meijer, F. Huber, A. van der Ploeg, S. Georgievska, V.T. van Hees.?Mcfly: Automated deep learning on time series.?SoftwareX, Volume 12, 2020.?doi: 10.1016/j.softx.2020.100548

This library simplifies the process of creating and testing models, you can test the code in this link: https://github.com/multiopti/MYWAI/blob/main/ts_class_cnn_mcfly.ipynb

You can put your comments here below, I would be happy to answer.

At?MYWAI?we promote agile, explainable, reliable and affordable ML at the edge.