SYNOPSIS:
CResult [options] Net Input
DESCRIPTION
The CResult command is used to evaluate the performance of classification
networks. A classification network has one output unit for each class
in a classification problem. The "choice" or "guess" of the network for
each input pattern is the class of the unit with the highest activity.
If no -m option is specified, all output units of the network are considered
class-units. The correct class for each pattern is defined as the class of
the unit with the highest target (see the Streams section).
The default output is percent patterns correct classified collapsed over
all classes and separately for each class. By default, the streams of the
network use 'Input' as the basename for data files (but see the -S option).
It is convenient to name the output units by their respective class. The
names of the units are used in the output of CResult.
OPTIONS
-S
This option changes the interpretation of the argument "Input". With the -S
option specified, the test data is taken from external files with the
basenames taken from the rows of the script file "Input". Otherwize "Input"
is itself a basename. In any case, the streams of the network read/write data
from files specified by the basename together with the information about file
extension and directory in each stream (see the Streams section).
-m object
Mark an object. By default all output units are used as class-units. This
options let you select a subset for of the units by specifying particular
units or groups of units. There can be any number of -m options on the
command line.
-c
Print a confusion matrix.
-n N
Print "within-top-N" statistics. This means that not only the patterns where
the correct class was ranked higest by the network are counted, but also the
patterns where the correct class was ranked as number two, three etc. 'N'
determines the lowest ranking to count.
-x label
Exclude frames were 'label' is the correct class.