NAME: NormGroup

SYNOPSIS

NormGroup [options] Net Input

DESCRIPTION

Normalizes the activities of the units of a group. This command works correctly only with linear units and its intended use is to normalize activites of units created by transforming the input unit's activations (see MakeCep and MakeDiff). The effect is analogous to that of the NormStream command, but in this case it is the connection weights that are adjusted to normalize the activities. Because the transformed activities are often treated as the input to the rest of the network and the connections of the transformation have plasticity 0.0, it makes sence to normalize them as if they where input units. 'Input' is a filename or with the -S option, an inputfile script. At least one -g option is necessary because there is no default behavior when no units are specified.

OPTIONS

-S This option changes the interpretation of the argument "Input". With the -S option specified, the training 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).

-g group Select the units in 'group' for normalization. Multiple -g options are allowed.

-d mult By defalt, NormGroup scans the training-data for the highest and the lowest value for each unit's activity. Then it sets up the inflowing connections so that these activities are mapped -1 and +1. This normalization is obviously sensitive to the extreme values and therefore it is sometimes better to use this option that normalizes based on the mean and standard deviation. "mean - mult * stddev" is mapped to -1 and "mean + mult * stddev" is mapped to +1.

-S	This option changes the interpretation of the argument "Input". With the -S option specified, the training 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).
-g group	Select the units in 'group' for normalization. Multiple -g options are allowed.
-d mult	By defalt, NormGroup scans the training-data for the highest and the lowest value for each unit's activity. Then it sets up the inflowing connections so that these activities are mapped -1 and +1. This normalization is obviously sensitive to the extreme values and therefore it is sometimes better to use this option that normalizes based on the mean and standard deviation. "mean - mult * stddev" is mapped to -1 and "mean + mult * stddev" is mapped to +1.