Learning DTW-Preserving Shapelets

Description

This code is used to learn Shapelet features from time series that form an embedding such that L2-norm in the Shapelet Transform space is close to DTW between original time series.

Usage

To learn a model and use it to perform $k$-means clustering in the Shapelet Transform space, one should run:

python clustering.py DatasetName [Conv]

Other implementations

A PyTorch implementation of the model is available at https://rtavenar.github.io/hdr/parts/02/shapelets_cnn.html#Learning-to-Mimic-a-Target-Distance.