manpages

isnns(1)

isnns(1) SNNS isnns(1)

NAME

isnns - interactiv, on the fly neural network training for SNNS

SYNOPSIS

isnns [ outpt_pattern_file ]

OPTIONS

outpt_pattern_file

A file with this name will be created and will log all training patterns used.

DESCRIPTION

isnns is a small program based on the SNNS kernel which allows stream-oriented network training. It is supposed to train a network with patterns that are generated on the fly by some other process. isnns does not support the whole SNNS functionality, it only offers some basic operations.

The idea of isnns is to provide a simple mechanism which allows to use an already trained network within another application, with the possibility to retrain this network during usage. This can not be done with networks created by snns2c. To use isnns effectively, another application should fork an isnns process and communicate with it over the standard input and standard output channels. Please refer to the common literature about UNIX processes, and how to use the fork(2) and exec(2) system calls.

COMMANDS

load net_file_name

This command loads the specified network into the SNNS kernel. After loading the network, the number of input units n and the number of output units m is printed to standard output.

save net_file_name

Save the network to the given file name.

prop i1 ... in

Propagates the given input pattern i1 ... in through the network and prints out the values of the output units of the network. The number of parameters n must match exactly the number of input units of the network.

train lr o1 ... om

Taking the current activation of the input units into account, this command performs one single training step based on the training function which is given in the network description. The first parameter lr to this function refers to the first training parameter of the learning function. This is usually the learning rate. All other learning parameters are implicitely set to 0.0.

learn lr i1 ... in o1 ...on

This command is nearly the same as a combination of prop and train. The only difference is, that it ensures that the input units are set to the given values i1 ... in and not read out of the current network. o1 ... on represents the training output and lr again refers to the first training parameter.

quit

Quit isnns after printing a final "ok>" prompt.

help

Print help information to standard error output.

SEE ALSO

The SNNS user manual under the SNNS installation directory or the hypertext manual at http://www-ra.informatik.uni-tuebingen.de/SNNS/UserManual/UserManual.html

snns(1) description of the whole package

fork(2) Creates a new process

exec(2) Executes a file

AUTHOR

Guenter W. Mamier