It looks like the DTW classifier picks a representative or exemplar training sample from each class, then does prediction based on the distance to each class's exemplar sample. It might be nice to specify an alternative strategy for prediction, e.g. KNN using DTW as the distance function, or taking the average distance to all the samples in each class.
For instance, in doing gesture recognition, it might be nice to be able to supply training samples from multiple people (or multiple styles from one person), and have the additional data do more than just (potentially) change the choice of exemplar for each class.
It looks like the DTW classifier picks a representative or exemplar training sample from each class, then does prediction based on the distance to each class's exemplar sample. It might be nice to specify an alternative strategy for prediction, e.g. KNN using DTW as the distance function, or taking the average distance to all the samples in each class.
For instance, in doing gesture recognition, it might be nice to be able to supply training samples from multiple people (or multiple styles from one person), and have the additional data do more than just (potentially) change the choice of exemplar for each class.