Click the Parameters button in the Analysis panel to show the KlustaWin parameters dialog box. This allows access to the following parameters.
Descriptions of the parameters are taken from the documentation for KlustaKwik.
Randomiser:
Variable sequence: Each time you run an analysis a different random initialisation occurs.
Fixed sequence: The same random initialisation sequence is used for each analysis run. This allows you to investigate the effects of changing other parameters.
BIC : AIC penalty ratio (default 0)
Amount of BIC (Bayesian information criterion) to use as penalty for more clusters. Default of 0 sets to use all AIC (Akaike information criterion).
Use 1.0 to use all BIC (this generally produces fewer clusters).
Minimum initial cluster number (default 2)
The random initial assignment will have no less than n clusters. The final number may be different, since clusters can be split or deleted during the course of the algorithm.
Maximum initial cluster number (default 10)
The random initial assignment will have no more than n clusters.
Distance threshold (default 6.91 = ln(1000))
Time-saving parameter. If a point has log likelihood more than d worse for a given class than for the best class, the log likelihood for that class is not recalculated. This saves an awful lot of time.
Full step every (default 10)
All log-likelihoods are recalculated every n steps (see Distance threshold).
Changed threshold fraction (default 0.05)
All log-likelihoods are recalculated if the fraction of instances changing class exceeds f (see Distance threshold).
Split steps (default 50)
Test to see if any clusters should be split every n steps. 0 means don't split.
Maximum iterations (default 500)
Don't try more than n iterations from any starting point.
Maximum possible clusters (default 100)
Cluster splitting can produce no more than n clusters. In practice it is best to be generous with this setting, because reducing the number to a low value does not always produce the “obvious” cluster allocations.