Package classifiers :: Module sim_cen_gauss_class

Module sim_cen_gauss_class

Simple centered Gaussian Classifier. It's designed for a case where there is a lot of data and where the data is a multivariate Gaussian.

Functions

float

logdet(cov)
Returns: the log of the determinant of a matrix.

source code

float

typ_trace(covlist)
This is used with boost_diag().

source code

boost_diag(covlist, diagboost)
This raises all the eigenvalues a little bit, so that processing can proceed even if the covariance is singular.

source code

go_auto(fd, Mu=False, coverage=3, ftest=0.25, cb=None, cfd_file=None)

source code

n_zero_cov_builder(d, N, modelchoice=None, trainingset_name=None, diagboost=1e-06)

source code

n_zero_cov_builder_drop(d, N, modelchoice=None, trainingset_name=None, diagboost=1e-06)
Needs to return a list, not an iterator.

source code

n_mu_cov_builder(d, N, modelchoice=None, trainingset_name=None, diagboost=1e-06)

source code

test1()

source code

Variables
	COVERAGE = `3`
	FTEST = `0.25`
	Fudge = `1e-06`
	PSYCO = `False`
	__package__ = `'classifiers'`

Imports: die, Num, dictops, fiatio, gpk_writer, QCR

Function Details

logdet(cov)

source code

Parameters:

cov (numpy.ndarray) - a 2-dimensional matrix

Returns: float

the log of the determinant of a matrix.

typ_trace(covlist)

source code

This is used with boost_diag(). It returns a typical value for the the trace of the covariance matrix, averaged over all matrices.

Parameters:

covlist (dict(str:numpy.ndarray)) - a dictionary mapping names to covariance matrices.

Returns: float

trace of covariance

boost_diag(covlist, diagboost)

source code

This raises all the eigenvalues a little bit, so that processing can proceed even if the covariance is singular. To first order, the fudge gets subtracted later on in the processing.

Parameters:

covlist (dict(str:numpy.ndarray)) - a dictionary mapping names to covariance matrices.
diagboost (float) - how much to boost

go_auto(fd, Mu=False, coverage=3, ftest=0.25, cb=None, cfd_file=None)

source code

Parameters:

cb (list(QCR.classifier)) - a function that builds the classifier. E.g. n_zero_cov_builder.
Mu (bool) - should the class means be allowed to be nonzero? Not here; Mu must be false.
cfd_file (str) - Pathname where the definition of the classifier should be written. None => don't write.