|
ZX Review #7-8-9-10
08 ноября 1997 |
|
reader-reader - With 'Light'. Spectrum and expert system.
The Reader - Reader
Music by ZET
(C) Group 'Light', Serpukhov
SPECTRUM AND EXPERT SYSTEM
We want to immediately respond to a question - why and why
needed and where it is used? A
most importantly - is it really possible to realize on the
Spectrum? Respond immediately - you can, even at
48K. But first, let's immediately
define what an expert
system. Under the expert system means a computer system that
can offer reasonable advice or carry out a reasonable solution
to the problem with some of the available source data.
An example of expert systems for
Spectrum can be, first and foremost, strategic games, and even
more chess programs. This assertion is based on the definition
of an expert system, which is endorsed by committee specialist
group on expert systems of the British Computer Society.
Of course, this article does
can not claim complete coverage of this section, but
We hope that it will, at
least piqued your interest
this region, which borders
artificial intelligence.
In this paper we consider an expert system which
suggests a human presence in the initial decision-making
process, ie, in learning the system.
Considered here an expert system can be simplified
represented as follows
algorithm:
Input
> Variables
v
Application
current
decisive
Rules
v
Assumption
the possible
outcome
All is good, but at an early stage in such an expert
system does not provide data on
the program could give
assumption about the possible outcome. And here on the stage by
a man who acts role of the teacher. Entered into the computer
those variables for which accurate
the outcome is known, teaches people
him. Thus, initially,
ie at the training stage, the algorithm
expert system might look as follows:
Input
> Variables <
v
Application
current
decisive
Rules
v
Assumption
the possible
outcome
v
Yes
This is true
v No
Modification
decisive
Rules
As can be seen from the algorithm, simple expert system is
easy to implement on the Spectrum. In proof of this on the
algorithm composes a program in BASIC, as a matter of common
language. If desired, this program is easy to translate into
any other language, even in assembler. For simplicity, assume
that there are only two possible outcomes: for example,
tomorrow it will rain tomorrow and will not be
rain. If you enter and run
this program, it asks the question about the number of
variables involved.
Number of the variable variable name
1 Pressure rises
2 Pressure drops
3 Wind is
4 No wind
Clearly, the choice of variables
lies entirely on your knowledge and
abilities. Then, in memory of the machine created three arrays.
Array R - must contain the rules of inference judgments (which
is developing a computer), the array V - values of the
variables presented in this case, the array V $ - used for
storage variable names.
The program asks for the names after the launch of these
variables. It also asks for the names of variables and O $ B $
- two possible outcomes. Further clarified details of a
specific example (for This program requires answers to yes /
no).
Having collected all the necessary information, the program
makes assumptions about possible outcome. When you agree with
the system, enter Y, or N - if you do not agree. Then move on
to another example. So way the process of
machine learning. It should be noted that the machine is
gradually improve their predictions about the possible outcome,
though how much better - depends on what information you have
entered at the beginning.
(Editor's note. - Training System will
much better if we introduce it REALLY
facts have occurred - and whether there was actually rain - and
not their agreement or disagreement with its decision).
And now he is listing the program.
10 CLS
20 INPUT "Enter the number of VARIABLES"; VAR
30 DIM V (VAR): DIM R (VAR): DIM V $ (VAR)
40 FOR I = 1 TO VAR
50 V (I) = 0
60 R (I) = 0
70 NEXT I: PRINT
80 PRINT "Name these variables'
90 FOR I = 1 TO VAR
100 INPUT "variable name:"; V $ (I)
110 NEXT I: PRINT
120 PRINT "Name the possible outcomes:"
130 INPUT "FIRST EXODUS:"; O $
140 INPUT "SECOND EXODUS:"; B $
150 PRINT
160 FOR I = 1 TO VAR
170 V (I) = 0
180 PRINT "VARIABLE:"; V $ (I)
190 INPUT "IS THIS VARIABLE <Y/N>"; A $
200 IF A $ = "Y" OR A $ = "y" THEN LET V (I) = 1
210 NEXT I
220 DC = 0
230 FOR I = 1 TO VAR
240 DC = DC + V (I) * R (I)
250 NEXT I
260 PRINT "possible outcome -";
261 IF DC> 0 THEN PRINT O $
262 IF DC <0 THEN PRINT B $
270 INPUT "It is true <Y/N>"; A $
280 IF A $ = "Y" OR A $ = "y" THEN GOTO 150
290 IF DC> 0 THEN FOR I = 1 TO VAR: R (I) =
R (I)-V (I): NEXT I
291 IF DC <0 THEN FOR I = 1 TO VAR: R (I) =
R (I) + V (I): NEXT I
300 GOTO 150
Once you decide
that have trained their expert system around the world, can be
safely deleted from the program training block (do not forget
only to keep the arrays with data) and to include this program
in part any more programs, or use it independently (eg for the
prediction of rain for days to come).
Imagine, however, that we
included the expert system into a strategic game, and
computer on the basis of this system manages a kind of fighting
robots having very little protection and the weakest means of
attack. We also assume that in our system is embedded rule
destroy all enemy targets within sight. And now
may come a time when
This weak robot heroically climb into your rocket launcher, and
as a consequence, about her and left. Here we have two choices:
the first - is to describe all possible situations (that not
very realistic) and the second - entered into the system the
opportunity oriented with minimal and
maximum values of the variables. In simple terms -
make sure that our hypothetical robot with the protection of,
say, in 1 unit is not useful in a fight
with the robot, with protection in 20
units. Then a block diagram of the
expert system will be as follows:
Home
<
v
Request
variables
v
Selecting the most
> Probable
outcome
v
Yes
Can set
nye minimal
nye and maxi
mum value
of influence
the choice of excitation
ble
outcomes
v No
DECISION
I must say that this
block diagram is not perfect, but due to the fact that
minimum and maximum values of the variables tested
only when necessary,
algorithm is obtained fairly
performance.
I must say that speed is not the last
role. Of course, when you build an expert system which
puts medical diagnoses, the
then performance is not
very crucial. Another thing - computer games. Here even
You can donate some intelligence for speed
program.
But the point here is this: There are two types of
decision-making - a parallel and consistent. Here's their
blokshemy:
Variables
v v v
v
Outcome
Parallel solution
Request variable <
v
Can I predict
some initial NO
v YES
Possible outcome
Consistent solution
Certainly consistent
solution is faster,
as the parallel solution
requires the input of all known
variables, and consistent
can conclude, using
only part of them. By the way,
there is speculation that it is
acts as the human brain
in decision-making.
There is no doubt that in general the parallel procedure to
obtain a more accurate solution than consistent. However, in
this case agree that we are more interested in speed of
calculation than correctness of the decision. Algorithm
calculation in this case is
as follows:
Home <
v
Request
variable
v
Selecting the most
> Probable
outcome
v
Yes
Can set
nye minimal
nye and maxi
mum value
expressions for the remaining
variables
affect
choice of excitation
can proceed
v No
DECISION
Thus, we have
fast algorithm, describing the action of an expert system in
reaching a decision.
Naturally, the material
This article does not describe all
possible solutions and algorithms for
this issue, because the problem of artificial intelligence has
not been solved until now. However, the simplest
algorithms can be applied even when programming on the
Spectrum. But implementation of these algorithms for you. So
what? You undertake Speccy create intelligent programs?
Those who liked this
idea, can write their thoughts on the following address:
142214 Serpukhov, ul.Podolskaya, d.107, kv.57
*
Other articles:
Similar articles:
В этот день... 21 November