from an Already-Clustered Dataset

Background about adding/deleting orders of clusters

You can use +addclust and +delclust to adjust the maximum number of clusters in (i.e. maximum order of) a dataset that you have already clustered with seismic cluster.

Purpose of adding clusters

After you run seismic cluster, you may wish that you had used a higher limit for the number of clusters (i.e. maximum order). You could increase the maximum order simply by rerunning seismic cluster using --force and a larger value for --max-clusters (-k). Since seismic cluster always begins with 1 cluster, you would need to repeat all the orders that you had already run before being able to add more – a waste of your computer resources. A more efficient command to increase the maximum order is seismic +addclust, which merely tests higher orders of clustering and, if successful, appends them to your existing results.

Purpose of deleting clusters

After you run seismic cluster, you may realize that higher orders are not useful for your analysis or have unwanted artifacts. If you simply ignored them, they would still waste space on your file system: higher orders result in larger files of cluster batches and cluster tables and more numerous files during graphing. You could decrease the maximum order simply by rerunning seismic cluster using --force and a smaller value for --max-clusters (-k). Since your clustering results would already exist – you would merely want to delete some of them – any approach that would require rerunning clustering would not be the most efficient. A more efficient command to decrease the maximum order is seismic +delclust, which merely deletes higher orders of clustering from your results.

How to add/delete orders to/from a Cluster dataset

Command line for adding orders

Suppose you have already run clustering with a maximum order of 3 (-k 3):

seismic cluster -k 3 {out}/{sample}/mask/{ref}/{sect}

If this dataset stopped at order 1 or 2 (because the BIC failed to decrease), then you cannot add any higher orders. If this dataset stopped at order 3 (because that was the limit), then you can add higher orders using seismic +addclust, specifying a new maximum order (e.g. 5) and the Cluster report(s) as input file(s):

seismic +addclust -k 5 {out}/{sample}/cluster/{ref}/{sect}

This command will start clustering with one more than the previous maximum order (in this case, the previous maximum was 3, so +addclust will begin at 4). The rules for stopping based on the BIC are the same as in seismic cluster (see Cluster setting: Maximum order (number of clusters) for more information). You do not need to specify any other settings for clustering, such as the number of runs or the threshold for convergence: +addclust automatically uses the same settings that you used originally in cluster.

You can give any number of Cluster report files as inputs for +addclust. See List Input Files for ways to list multiple files.

The only new files produced are mus and props files for each additional clustering run. The Cluster report file, the batch files, and the counts file are all updated in-place, as if you had run cluster the first time with the new -k.

Command line for deleting orders

Suppose you have already run clustering with a maximum order of 3 (-k 3):

seismic cluster -k 3 {out}/{sample}/mask/{ref}/{sect}

You can delete orders above a limit using seismic +delclust, specifying a new maximum order (e.g. 2) and the Cluster report(s) as input file(s):

seismic +delclust -k 2 {out}/{sample}/cluster/{ref}/{sect}

This command will simply delete the results for higher orders (in this case, 3) from the Cluster report file, the batch files, and the counts file. It does not delete the mus and props files, since these files are small, not used by any subsequent step, and not constrained by the maximum order.

You can give any number of Cluster report files as inputs for +addclust. See List Input Files for ways to list multiple files.

The Cluster report file, the batch files, and the counts file are all updated in-place, as if you had run cluster the first time with the new -k.

Note

Be careful with +delclust because it will delete all orders greater than max-clusters (-k). Deleted orders cannot be restored directly; to regenerate them, you would need to rerun clustering with +addclust.