This adds motif features, which randomly selects 64ish motifs from the dataset, and runs a matrix profile join against each incoming sample. We use the min, median, and max of the resulting joins as a feature. This adds a significant amount of overhead to the training/inference process. It'd be nice to optimize the matrix profile algorithm so it runs faster.
This adds motif features, which randomly selects 64ish motifs from the dataset, and runs a matrix profile join against each incoming sample. We use the min, median, and max of the resulting joins as a feature. This adds a significant amount of overhead to the training/inference process. It'd be nice to optimize the matrix profile algorithm so it runs faster.