NAME
set_em_cluster_options - set_em_cluster_options
SYNOPSIS
#include "r/r_cluster.h"
Example compile flags (system dependent):
-DLINUX_X86_64 -DLINUX_X86_64_OPTERON -DGNU_COMPILER
-I/home/kobus/include
-L/home/kobus/misc/load/linux_x86_64_opteron -L/usr/lib/x86_64-linux-gnu
-lKJB -lfftw3 -lgsl -lgslcblas -ljpeg -lSVM -lstdc++ -lpthread -lSLATEC -lg2c -lacml -lacml_mv -lblas -lg2c -lncursesw
int set_em_cluster_options
(
const char *option,
const char *value
);
DESCRIPTION
Cluster data perturbation.
cluster-discrete-threshold
Cluster threshold for excluding discrete probabilities.
cluster-num-tries-per-cluster-count
When trying different cluster counts, this option specifies the number number of EM tries each number of clusters gets (must be at least one) (default 1)
cluster-num-cluster-count-samples
Number of log-spaced EM clusters counts to try (must be at least one) (default 30).
cluster-max-num-iterations
The maximum number of EM iterations to use (default 20).
cluster-iteration-tolerance
EM iteration tolerance (default 1e-6).
cluster-plot-log-likelihood
Plot EM log likelihood vs num clusters. (Boolean, default false)
cluster-norm-data
Normalize Cluster data. (boolean, default false)
cluster-var-offset
Cluster variance offset (default 1e-4)
cluster-use-unbiased-var-estimate-in-M-step
Unbiased estimate for variance %s used in the M step. (Boolean, default false)
cluster-held-out-data-fraction
Held out data fraction.3e. (default 1/10)
cluster-tie-var
Use cluster tie variance (Boolean, default false)
cluster-tie-feature-var
Use cluster tie feature variance (Boolean, default false)
cluster-tie-cluster-var
Use cluster tie cluster variance. (Boolean, default false)
cluster-model-selection-training-MDL
Use training data MDL criterion for model selection. (Boolean, default false)
cluster-model-selection-held-out-LL
Use held out data log likelihood criterion for model selection. (Boolean, default True)
cluster-model-selection-held-out-MDL
Use held out data MDL criterion for model selection. (Boolean, default false)
cluster-model-selection-held-out-corr-diff
Use held out data correlation difference criterion for model selection. (Boolean, default true)
cluster-model-selection-held-out-max-membership
Use held out data maximum membership criterion for model selection. (Boolean, default true).
cluster-EM-stop-criterion-training-LL
Use training data log likelihood as EM stopping criterion. (Boolean, default true).
cluster-EM-stop-criterion-held-out-LL
Use held out data log likelihood as EM stopping criterion. (Boolean, default false).
cluster-use-initialized-cluster-means-variances-and-priors
Use initialized cluster means and variances in the EM algorithm. (Boolean, default false).
cluster-max-num-CEM-iterations
The maximum of N CEM iterations are used. (default 100)
cluster-write-CEM-intermediate-results
Whether to write CEM intermediate results. (default true)
cluster-force-equal-prob-for-CEM-split-and-merge
Whether the probabilities of choosing a CEM split or a merge operation are forced to be equal. (default true)
cluster-crop-feature-dimensions
Crop some feature dimensions before clustering (default false)
cluster-crop-num-feature-dimensions-left
Number of feature dimensions to the left to crop. (default 0)
cluster-crop-num-feature-dimensions-right
Number of feature dimensions to the right to crop. (default 0)
SEE ALSO
get_independent_GMM
,
get_independent_GMM_using_CEM
,
get_independent_GMM_with_shift
,
get_independent_GMM_with_shift_2
,
get_GMM_blk_compound_sym_cov
,
get_GMM_blk_compound_sym_cov_1