TESTN

Error estimate of discriminant for normal distribution.

    E = TESTN(W,U,G,N)

Input
W	Trained classifier mapping
U	C x K dataset with C class means, labels and priors (default: [0 .. 0])
G	K x K x C matrix with C class covariance matrices (default: identity)
N	Number of test examples (default 10000)

Output
E	Estimated error

Description

This routine estimates as good as possible the classification error  of Gaussian distributed problems with known means and covariances. N normally distributed data vectors with means, labels and prior  probabilities defined by the dataset U (size [C,K]) and covariance  matrices G (size [K,K,C]) are generated with the specified labels and are  tested against the discriminant W. The fraction of incorrectly classified  data vectors is returned. If W is a linear 2-class discriminant and N is not specified, the error is computed analytically.

See also

mappings, datasets, qdc, nbayesc, testc,