This is a supporting page to our paper -
Logical-Shapelets: An Expressive Primitive for Time
Series Classification

by

Abdullah Mueen, Eamonn Keogh and Neal Young
The paper is
here.

Code and Executables 

We have two versions of our algorithm.

 

1. Shapelet: The fast version of the original shapelet algorithm by using the speedup techniques described in the paper.

2.   LogicalShapelet: The logical shapelet version that searches for multiple shapelets for more expressive concepts for the training dataset.

Both of the versions are available here.

 

Spreadsheet of Experimental Results 

We have compiled results of all the experiments in a spreadsheet

Few things to note ...

  • All time measurements are in seconds
  • All the times are in seconds unless specified otherwise
  • All the distances are in z-normalized space.
  • All the codes use "Early Abandoning".

Datasets

  • For scalability experiments, we use 24 training datasets to check the speedups. Each row in every file is a time series and the first value of a row is the class label.
  • Length of the time series within and across datasets are not fixed.
  • The datasets are : CBF, FaceFour, Trace, GunPoint, OSU Leaf, Lightning2, ECG, Plane, Car, Coffee, Olive Oil, Beef, Symbols, Diatom, Motes, ECGFiveDay, SonyAIBO, Haptics45, ItalyPowerDemand, Chlorine, TwoPat, Cricket, Arrowhead and BirdSong.

CRICKET: Automatic Scorer

The training set. (9 time series)

The testing set.  (98 time series)

The new testing set. (64 time series) This is generated in our lab using a smart phone's accelerometer.

For all sets, the label 0 is for wideball and label 1 is for no ball.

c

Sony AIBO: Surface Detection

The training set. (20 time series)

The testing set. (601 time series)

sony
Passgraphs: Preventing Shoulder-Surfers
The training set. (69 time series)
The testing set. (131 time series)
passgraph

This page is created by -
Abdullah Mueen
Department of Computer Science and Engineering,
University of California - Riverside.