K-Spectral Centroid: Cluster Time Series by Shape
Codes information
The K-Spectral Centroid algorithm clusters time series by their shape, and finds the most representative shape (the cluster centroid) for each cluster. We provide the simple version of the K-SC code for Matlab. For brevity, this version does not include the Haar-Wavelet based incremental implemention discussed in the paper.
| Function description | |
|---|---|
| test_ksc.m | Script that tests K-SC |
| ksc_toy.m | K-SC Matlab implementation code |
Source (citation)
- J. Yang, J. Leskovec. Temporal Variation in Online Media. ACM International Conference on Web Search and Data Mining (WSDM '11), 2011.
Files
| File | Description |
|---|---|
| ksc_toy.zip | K-SC Matlab Version |
Usage
[mem cent] = ksc_toy(A,K)
Description
- A: A set of time series. Each row vector A(i,:) corresponds to each time series
- K: The number of clusters (K=6 in the paper)
- mem: Membership for each time series. mem(i) = the cluster index that time series i belongs to
- cent: A set of cluster centroids. Each row vector cent(i,:) corresponds to each cluster centroid
Demo
Unpack the ksc code file in the directory with the data files, and run the following.
test_ksc(1) %demo for Memetracker phrases
test_ksc(2) %demo for Twitter Hashtags