spr_calc_lda.c File Reference

calculate the lda transformation for a given preprocessing and segmentation file (classes). More...

Detailed Description

calculate the lda transformation for a given preprocessing and segmentation file (classes).

spr_calc_lda [-DIAG](flag: ) [-PCA](flag: ) [-LDA](flag: ) [-EA](flag: ) [-stats stat_file]
    [-add add_stat_file] [-o matrix] [-mvg mvg_file] [-s segfname] [-range b:e](0:-1)
    [-units units] [-cgs classes/gaussians/states]
    [-ssp signal processing script](SPR_BSS_DEV_NULL) [-obs obsdir] [-suffix suffix](trk)
    [-Alpha class weighting](1.0) [-Beta class weighting](1.0) [-clip clipping]

Parameters

-DIAGflag	calculate diagonal covariance matrices only.
-PCAflag	calculate the PCA transformation on the average within class covariance matrix.
-LDAflag	calculate the LDA transformation (default action).
-EAflag	error analysis on existing models (full covariance).
-stats<em>stat_file</em><a	name="spr_calc_lda.stats" class="el"> Write the class statistics (Npoints, Moment1, Moment2) to the given file. The values Moment2(i,i-j) are stored for j=0..n, i=j..n, with i the fast running and j the slow running index. So first the diagonal, then the first lower diagonal, ...
-add<em>add_stat_file</em><a	name="spr_calc_lda.add" class="el"> Read (additional) class statistics. Multiple file can be specified by seperating by ' .AND. '.
-o<em>matrix</em><a	name="spr_calc_lda.o" class="el"> The resulting LDA/PCA-matrix.
-mvg<em>mvg_file</em><a	name="spr_calc_lda.mvg" class="el"> Write out the diagonal covariance gaussian set.
-s<em>segfname</em><a	name="spr_calc_lda.s" class="el"> The state-based segmentation file.
-range<em>b:e</em><a	name="spr_calc_lda.range" class="el"> Optional begin- and end-entry in the segmentation file. Counting starts at 0.
-units<em>units</em><a	name="spr_calc_lda.units" class="el"> The units file (.arcd or .cd format).
-cgs<em>classes/gaussians/states</em><a	name="spr_calc_lda.cgs" class="el"> File describing the class/gaussian/state relation. Format: [name] <name> [classes] <class_1> ... <class_N> [gauss2class] <class_for_gauss1> ['REP' <number>] ... [state2class] <class_for_state1> ['REP' <namber>] ...
-ssp<em>signal	processing script The signal processing script used to preprocess the input data.
-obs<em>obsdir</em><a	name="spr_calc_lda.obs" class="el"> Observation directory.
-suffix<em>suffix</em><a	name="spr_calc_lda.suffix" class="el"> File suffix used for all entries in the segmenation file.
-Alpha<em>class	weighting Non equal weighting of the points in the classes for the within class covariance matrix. Alpha < 1.0: give more importance to the classes with few points. Alpha > 1.0: give more importance to the classes with lots of points.
-Beta<em>class	weighting Non equal weighting of the points in the classes for the between class covariance matrix. Beta < 1.0: give more importance to the classes with few points. Beta > 1.0: give more importance to the classes with lots of points.
-clip<em>clipping</em><a	name="spr_calc_lda.clip" class="el"> Clip moment 0 (number of occurances); clipping is done right after aggregating the statistics; the clip value (upper limit) can be a simple numerical value <N>, a fraction (%<N>), a number of classes (#<N>), or a file containing class specific clip values; in case of '' or '#', the maximum is equal to the number of times the <n>'th class occured, with <n> expressed as a fraction of the total number of classes (''), or just an index ('#'); the classes are ordered from large to small.

Calculate the lda transformation for a given preprocessing and segmentation file (classes). Calculate the PCA/LDA transformation for a given preprocessing and segmentation file (classes) or do error anlysis (classification error and perplexity) of an existing model.

Author

Kris Demuynck

Date

22/05/97

Revision History:

12/09/97 - K.D.

creation

02/02/99 - K.D.

added error analysis

03/03/11 - K.D.

can now also make mvg-files and work with diag-cov

10/06/18 - K.D.

read fvg files + clip class weights (M0)