SHORT DOCUMENT FOR USING “CL”

Short Document for Using CL

What is CL?
About CLNG.
Getting Start with CL.
Commands Useful for the Novice.
Sample Session.
Reference.

What is CL:

CL, stands for Competitive Learning, is a command-driven system designed by McClelland and Rumelhart for training and testing neural networks using competitive learning topology. Competitive learning paradigm is one of the simplest paradigms in unsupervised learning and has been used in a number of applications, such as speech perception, word recognition, visual perception, and cell formation.

About CLNG:

Preparing data files to run CL is a complicated and time consuming process. To simplify the task, a program generator called CLNG is developed to generate all files needed to run CL. However, CLNG is designed for solving cell formation problem only. The machine/part matrix, which is needed for cell formation, must be stored in a file. The format of this data file must be organized as follows:

R (number of rows - machines)
C (number of columns - parts)
Machine/Part Matrix (R x C binary matrix)

An sample file "Exa" is stored in the server for reference, where the machine/part matrix contains 12 machines and 10 parts.

Getting Start with CL:

The "CL" program and its utility "CLNG" were stored in my course server under the directory of "m528\systems\cl". First, create a directory (e.g., CL) in your hard drive, download the files into that directory. To access the programs, you must quit the Window environment to DOS environment. Change your default directory to where CL was stored. You must run "CLNG" to generate the necessary files for use by CL and then run CL by typing "CL file.tem file.str", where "file" is the corresponding file name you designed in the CLNG run.

Commands Useful for the Novice:

CL is a command-driven system; that is you must enter proper command to trigger the system. The following commands are useful for the novice:

strain: to train the network using a fixed sequential order of training data.

ptrain: to train the network using a permuted (random) order of training data.

tall: to test all patterns on the pattern list one at a time.

test: to present a pattern from the pattern list for testing.

newstart: to allow the program to be retested on the same problem, but with a new series of random number.

clear: to clear the display area.

reset: to reset activation of units in the network to resting level.

set/: to examine the current value of a variable and to set it to a new value if desired.

run: to pass a command to the command interpreter under which you are running the simulation program.

quit: to exit the program.

Sample Session:

The following sequence of commands (in bold face) can be used to setup the initial weights (reset or newstart), train the network (ptrain or strain), and test the network (tall or test):

reset (to reset initial weights)
ptrain (train the network in random order)
(Repeat the ptrain command until no change in the weights)
tall (test the network for all units)
(Write down the result. Press "Enter" key until all units have been tested)

Note:

Use "newstart" instead of "reset" if you would like to change to a different set of initial weights.
Use "strain" instead of "ptrain" if you would like to train the network in sequential order.
Use "test" instead of "tall" if you would like to test each individual unit.

Reference:

McClelland, J. L. and Rumelhart, D. E., Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises, The MIT Press, MA., 1988.