The Complete SPRLIB & ANNLIB

net_perf

- compute the performance of a network on a dataset

SYNOPSIS

ERROR_STRUCT *net_perf (net, dset, err_struct, err_threshold, options)

ARGUMENTS

NET *net A pointer to a feedforward NET.
DATASET *dset A pointer to a DATASET.
ERROR_STRUCT *err_struct A pointer to an ERROR_STRUCT structure containing the errors. If NULL, the routine allocates a new structure.
double err_threshold Value below which the difference between the target and actual output is considered to be low enough to find the classification correct.
long options An option parameter of type PERF-flags.

RETURNS

A pointer to an ERROR_STRUCT of the network in case of success, NULL in case an error was encountered.

FUNCTION

This function evaluates the errors for each sample in dset using the sample_perf function. If option is set to LONG_PERF, these values are stored for each sample. In all cases, the overall ERROR_STRUCT is calculated for the complete dset.

NOTE

The dataset dset must be an IOSET and a LEARNSET (see DATASET-flags). If option = LONG_PERF, the perf_init function must be called before net_perf can be called. The ERROR_STRUCT 's AllCorrect flag now indicates whether or not all output units were considered to be correct for all samples in dset. The values of MinError and MaxError are taken over the complete dset, as is PercentageCorrect. The MeanSquaredError is divided by the number of samples in dset. Finally, the PercentageClassCorrect field indicates which percentage of the samples in dset is classified correctly (for the definition of correct classification, see sample_perf.

SEE ALSO

perf_init, sample_perf, perf_free, error_struct_array

This document was generated using api2html on Thu Mar 5 09:00:00 MET DST 1998