FAQs‎ > ‎

3 - Segmentation


Why do the number of maps and number of clusters differ?

This actually depends on your parameters. First remember that the segmentation is mainly two steps: a mathematical clustering, followed by some temporal processings.

The first step of the Segmentation gives us "clusters", and they are the right amount, so requesting 10 clusters will yield, well, 10 clusters...

Now the second step (temporal processings) will alter this state of affairs: requesting a Sequentialization of clusters will give us more segments than clusters, and conversely rejecting small segments will remove some. Combining both trends, the final number of segments is most of the time different from the number of clusters...

Note: there is a kind of semantic difference between a "cluster" and a "segment". A cluster is something out of the Clustering algorithm, a segment is a cluster with some temporal and physiological meaning.


What are these cryptic numbers in the segmentation file names, like ".08.(12).seg"?

This is the consequence of the previous point: the number of requested clusters and the number of final segments might differ according to your parameters. So the first number, here .08, stands for the number of clusters, and the second number, here .(12), stands for the resulting number of segments.

 

Why I get a message "Not enough TFs for the requested amount of clusters" 

You need a minimum amount of samples (read: Time Frames) to run the segmentation:

  • The Clustering itself needs at least n TFs for a maximum of n clusters.
  • The computation of the KL quality measure needs one neighbor on each side, therefore for n clusters, you need at least n + 1 TFs.
  • Plus, Cartool filters the W curves to compute the KL, and it also needs 1 more neighbor to do this.

We end up with these rules:

  • Max. requested segments <= Total Number of TFs
        and
  • ( Max. requested segments - Min. requested segments ) + 4 < Total Number of TFs

(Well, there also are some side effects if  (Min. requested segments) is 1, but Cartool will tell you the real limits anyway).

 

Fitting Maps: How do I specify groups with different number of subjects / files?

You may have different number of files in your groups, f.ex. 50 subjects and 10 patients. But currently, Cartool does not allow you to fit groups of different sizes, as the output matrix of variables would be missing some parts.

You can however bypass this limitation by doing the following steps:

  • Fit 1 group at a time
  • In the Statistics dialog, drop every .scv files generated by the above fittings
  • Check you have all your groups together, then don't forget not to use paired tests, but only unpaired!

You can actually fit together groups of the same size: f.ex. for 4 groups with 50, 10, 50 and 10 files, you can fit groups 1 and 3 together, then groups 2 and 4.

Comments