ClassAverages

class_averages

Purpose

This program allows you to divide a selfile into the desired number of classes. Vector quantization with correntropy and a probabilistic criterion is used for creating the subdivisions. The program produces an output directory structure that can be easily processed by SortImages.

Usage

$ mpi_class_averages ...

Parameters

• `` Set of images to divide
• -o [root"class"] = Name of the output classes. You can provide an existing directory as-o directory/rootname.
• -iter [N20] = Maximum number of iterations at each level of subdivisions
• -codes0 [N2] = Number of code vectors in the first iteration
• `` Selfile with the initial code vectors
• -codes [N16] = Final number of code vectors. It does not need to be a power of 2
• -neigh [N3] = Number of neighbors of each codevector to consider
• -minSize [percentage20] (or the user supplied value) than the total number of images divided by the number of codevectors at that level, the codevector is considered to represent too few input images and it is removed. The largest codevector at that moment is split.
• `` Do not check mirrors when comparing input images to codevectors.
• `` Use correlation instead of correntropy
• `` Use classical multireference alignment (maximum of the distance) instead of the enhanced clustering criterion based on the likelihood of belonging to a class compared to the likelihood of not-belonging to that class
• `` Align original images at the end so that all of them are aligned with their respective class
• `` Perform a fast estimate of the probability density functions. This option is suboptimal bu usually gives good results

Examples and Notes

Compute 128 classes using 8 processors

$ mkdir Classes
$ mpirun -np 8 `which xmipp_mpi_class_averages` -i g0ta.sel -codes 128 -o Classes/class -fast
$ xmipp_sort_images -i Classes/class.sel -oroot Classes/class_sorted -processSelfiles

The result is stored inClasses/class.sel, there are many intermediate selfiles with the different groups at different levels. For each codevector, there is a list of the images assigned to that class.

USER's COMMENTS

--Main.CoSS - 25 Jun 2009