NeurDS(1) NeurDS(1)
31 May 2000
NAME NeurDS - Neural Design and
INVOKING:
SYNTAX
% model [ifile [ofile]]
where: ifile = input script file (described below),
default is TT: ofile = output trace file, default is TT:
------------------------------------------------------------
Note that INPUTs and OUTPUTs are specified in HEXADECIMAL, where each
bit signifies a UNIT in the INPUT or OUTPUT layer (lowest bit = UNIT
0). For example an INPUT of 1F would turn on UNITs 0 through 4 in the
input layer (with all other UNITs off).
-------------------------------------------------------------
DESCRIPTION
A general purpose tool for building, running and analysing Neural
Network Models in an efficient manner
COMMANDS
The NeurDS simulator has the following commands:
-------------------------------------------------------------
; - comment everything until end of line
use: Documentation of a command script
params: everything until the end of line
defaults: <none>
example: ; This is a comment
-------------------------------------------------------------
<num> - load a pattern into memory
use: load patter for use by ASSOC or INPUT commands
params:
index - array index for pattern (>= 0 and < 5000)
inpstr - input string of HEX data outstr - output
string of HEX data
defaults:index - <none> inpstr - <none> outstr -
<none>
- 1 - Formatted: November 21, 2024
NeurDS(1) NeurDS(1)
31 May 2000
example: 123 13ABCF0 2D7
-------------------------------------------------------------
ASSOC - associate inputs with outputs (learn)
use:Train the system
params:inpmin - first pattern index to use inpcnt -
number of patterns to use maxloop - maximum number of cycles to
run for (unless desired accuracy reached first, see PARAMS
command) canrand - if no progress is being made, is the system
allowed to randomize the network force - force
randomization after how many loops? tryrand - randomize after
this many loops of not < tryerr tryerr - error level for
tryrand
defaults:inpmin - 0 inpcnt - (lastindex) maxloop -
999999 canrand - 0 (no) force - 999999
tryrand - 999999 tryerr - 999999.999
example: assoc 0 500 999 1 200 100 0.1
-------------------------------------------------------------
CLEAR- delete all UNIT and LINK
use:remove all network definitions
params:<none>
defaults:<none>
example: clear
-------------------------------------------------------------
DUMP - output information from the network
use:Get different levels of status information
params:level - 0 = output value for current input - 1
= activation values of all layers - 2 = full dump of all
neuron states - 3 = full dump of all neurons and links
- 4 = screen dump from window package
defaults:level - <none>
example: dump 3
-------------------------------------------------------------
- 2 - Formatted: November 21, 2024
NeurDS(1) NeurDS(1)
31 May 2000
EXIT - leave the program
use:ignore rest of script, and stop program
params:<none>
defaults:<none>
example: exit
-------------------------------------------------------------
HELP - get command help
use:type out this file
params:<none>
defaults:<none>
example: help
-------------------------------------------------------------
INPUT - ask for network to run
use:Try using a network after its created (see ASSOC)
params:inpmin - input pattern index to start at inpcnt -
input pattern count
defaults:inpmin - 0 inpcnt - last input pattern index
example: input 500 500
-------------------------------------------------------------
LINK - connect two UNITs together
use:connect 2 UNITs (from lower to higher).
params:fl,fu - FROM layer , FROM unit (numbering based at 0)
tl,tu - TO layer , TO unit (numbering based at 0)
weight - initial weight value
defaults:fl,fu - <none> tl,tu - <none> weight -
randomly generated
example: LINK 1,0 2,0 0.3 0.01 LINK 1,0 2 (connect to
- 3 - Formatted: November 21, 2024
NeurDS(1) NeurDS(1)
31 May 2000
everyone on level 2) LINK 1 2 (connect from/to everyone on levels
1/2)
-------------------------------------------------------------
PARAMS - set global parameters
use:Set overall system parameters
params:A1...A5- values of model specific parameters
defaults:A1...A5- model specific
example: params .5 .9 .1 0 0
-------------------------------------------------------------
RANDOM - randomize the network
use:randomize link values and neuron states
params:<none>
defaults:<none>
example: random
-------------------------------------------------------------
SCRIPT - read from a script file
use:allow command nesting via scripts
params:file - name of file to use as script
defaults:file - <none>
example: script xor.script
-------------------------------------------------------------
TRACE - create a trace (log) file
use:allow a log file to be started/ended
params:file - name of file to use as trace (log)
defaults:file - stop trace
example: trace xor.log (start log)
- 4 - Formatted: November 21, 2024
NeurDS(1) NeurDS(1)
31 May 2000
race (end log)
-------------------------------------------------------------
UNIT - create a new processing element
use:Create units for a network layer
params:lev,num- level and number of units
defaults:lev,num- <none>
example:UNIT 1,10 UNIT 2,5 UNIT 3,2* (* == last layer ==
output layer)
-------------------------------------------------------------
VERBOSE - turn on/off verbosity flags
use:Get more or less info out during a run
params:echo - Echo input lines report - number of loops
between reports numplot - number of loops between plots
adump - report after each pattern (not just summary)
pdump - plot after each pattern (not just summary)
showdiff- plot histogram in window mode mix - randomize
patterns during learning
defaults:echo - 1 report - 1 numplot - 1
adump - 0 pdump - 0 showdiff- 0 mix - 0
example: verbose 1 10 1 0 0 1 0
-------------------------------------------------------------
AUTHOR
DBW - David B. Wecker, Database Systems Research Group
MAN PAGE
Rush Job!!
- 5 - Formatted: November 21, 2024
