OPT - A simple probabilistic inference engine the PC


OPT is a simple probabilistic inference demonstration for the IBM PC. It is
useful for on-line documentation, CAI etc., It is provided with
full source code to encourage experimentation.

Startup

1. Boot the system.
2. Insert the disk containing OPT.COM in Drive A:
3. If the DOS prompt is not A>, type A: and press the ENTER key to switch
to drive A:.
4. Type OPT filename and press RETURN. "filename" is the name of
file containing the marked text. For example, "OPT test.rul"
will analyze the sample file which was supplied with OPT.
5. OPT will display its opening screen and wait for you to press
a key. If you did not supply a filename on the command line, OPT
will prompt you for one.
6. HYPE will iterate and display the initial soloution one page at a
time. The initial probabilities will be displayed in reverse video.
The probabilities will be displayed scaled from 0 to 100.One threaded
concept will be diaplayed with a special cursor called the button. On
a color display the button is yellow text on a black background. On a
mono screen the text under the button is underlined. You can move
through the text with the following commands:

UP - up arrow - move to next highlighted area
Down - down arrow - move back to previous highlighted
area
Esc - exit the current window and begin the optimization process
to find a new solution.
PgUp - display the next page of text.
PgDn - display the previous page
Enter - fix the the probability at the current value. Fixed values
have a small diamond displayed to the left of the name. Pressing
Enter on an item which is fixed, releases that value.
0 .. 9 - change the number.
del - delete the character under the cursor.
-> - right arrow - move right one space.
<- - left arrow - move left one space.
<= - backspace - delete previous character.
F10 - quit the program and return to DOS

Fixing a value for a probability is equaivalent to entering new information
into the system. Entering 100 for a value is euivalent to saying that
that hypothesis is definitely true. 0 indicates that a hypothesis is
definitely false. Anything in-between indicates the degree of
certainty that you judge that parameter to have. Allowing a value to
float by not fixing it means that you do not wish to enter naything
for that value.

Pressing the key causes OPT to try to find a new soloution to
the network. As OPT runs you will see a display of the interation
count, the object function value and the difference between successive
steps. The difference should always be negative. The objective value
will be quite large if the solution is infeasible. Eventually, it
should decrease to a relatively small value. Pressing a key will stop
the optimization process and display the probabilities. At this point
you can change values or press to continue the search process.
Changing a probaility value, but not fixing it should have no effect
on the outcome of the optimization process, but may increase or
decrease the number of iterations.

OPT rule files

Text files for OPT should be ASCII files which contain no
control characters. The file contains rules and default values. At
present only implication, AND and mutual exclusion rules are
implemented.

Implication and AND
These rules should be entered first. Rules have the following syntax:

hypothesis,w = hypothesis1,hypothesis2, .... .

The terminating period is necessary. This rule says

hypothesis1 and hypothesis2 and ... implies hypothesis with weight w.

w is a number between 0 and 1.0. If it is absent, a value of 1 is
assumed.

Mutual Exclusion
Mutually exclusive probabilities are entered as:

=hyp1,hyp2 .... .

The terminating period is necessary.

Defaults
After all the rules have been entered, the default values are
enterend. Begin the table of defaults with the word %default. Defaults
are entered as:

hyp = value .

value must be >=0 and <=1. Any hypothesis not given a default is
assignes a default of 0.

Hypothesis names may be up to 60 characters long and may not contain
spaces.

Lines beginning with a ; are condsidered comments.

The following is the contents of the file test.rul, supplied with OPT.

; Rules

survival,0.25 = cadaver,low_matching_factor,no_complication .
survival,0.8 = living,no_complication.
survival,0.4 = cadaver,high_matching_factor.

; Mutual Exclusions

=high_matching_factor,low_matching_factor.
=cadaver,living.

%defaults
survival = 0.25 .
cadaver = 0.75 .
living = 0.25 .
low_matching_factor = 0.75 .
high_matching_factor = 0.25 .
no_complication = 0.4 .




Good luck with OPT. We would be very interested in hearing of
your experiments, enhancements or even (gasp) bugs that you may find.
Please write to us with your comments or questions.

Bill and Bev Thompson
C/O AI Expert Magazine
500 Howard St.
San Francisco, CA 94105

or on the AI Expert BBS on Compuserv. Our id is BillandBev Thompson,
[76703,4324]. You can also contact us on BIX, our id is bbt.

Bill and Bev Thompson