1) Read table generated by trend labeller.
2) Normalize time series in each row (I can think of 3 or 4 reasonable ways to do this. The paper probably has something specific in mind)
3) Cluster the normalized time series with lablel=1. Cluster the normalized time series with lablel=0. k-means with cluster coherence diagnostics to select k for each set (Scree plot, etc)
4) Write the cluster centroids to a new DB table with schema (ClusterID, label, t1, t2, .... tn)
1) Read table generated by trend labeller. 2) Normalize time series in each row (I can think of 3 or 4 reasonable ways to do this. The paper probably has something specific in mind) 3) Cluster the normalized time series with lablel=1. Cluster the normalized time series with lablel=0. k-means with cluster coherence diagnostics to select k for each set (Scree plot, etc) 4) Write the cluster centroids to a new DB table with schema (ClusterID, label, t1, t2, .... tn)