# Fuzzy relations vs. Mamdani model (continued)

From: Pim van den Broek (pimvdb@cs.utwente.nl)
Date: Sun Dec 31 2000 - 04:36:01 MET

• Next message: P. Sarma: "Re: Fuzzy Control"

--------------63D9D56CE6B0D8F53C5FA784
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Hello,

A couple of weeks ago there was a discussion in this group following a
question of Andrej Albert on the difference between two models of
approximate reasoning. Continuing this discussion I will consider a small
example, and show that the choice of the model should be dependent on
the interpretation of the rules, and moreover, that for some interpretations
of the rules neither of the two models is appropriate.

Suppose we have a set of rules of the form "IF A THEN B" where A is a set
of symptoms and B is a set of diseases. The intention of the rules is to obtain
a diagnosis of the disease(s) of a patient, given information about the symptoms
this patient exhibits. We will only consider crisp sets here, taking the point of

view that a model for approximate reasoning should produce reasonable results
also when only crisp sets are considered.
Given a rule "IF A THEN B" where A is a set of symptoms and B is a set of
diseases and also given a set A' of symptoms, the resulting set B' of diseases
will depend on the meaning of the rules. A rule might be applicable if the
patient
exhibits ANY of the symtoms, or if the patient exhibits ALL the symtoms.
The result of the rule might mean that the patient suffers from ALL diseases from

the set, or that the disease is ANY of the diseases of the set.
This leads to four possible interpretations of a rule "IF A THEN B":

Interpretation 1 : IF the patient exhibits all symptoms from A THEN the patient
has
all diseases of set B
Interpretation 2 : IF the patient exhibits all symptoms from A THEN the disease
of this patient belongs to set B
Interpretation 3 : IF the patient exhibits anyl symptoms from A THEN the patient
has
all diseases of set B
Interpretation 4 : IF the patient exhibits any symptoms from A THEN the disease
of this patient belongs to set B

Now consider the case where there are 3 symtoms and 3 diseases, and
symptoms and diseases both are denoted by the numbers 1,2 and 3.

Consider the rule : IF {1,2} THEN {1}
We consider two input sets of symptoms: {1} and {3}.
The following table contains for both input sets and all four interpretations of
the
rule the desired output set of diseases.

Int. 1 Int. 2 Int. 3 Int. 4
Input {1} | {} {1,2,3} {1} {1}
Input {2} | {} {1,2,3} {} {1,2,3}

This table is computed as follows:
The desired result is {1} ( the right hand-side of the rule), in case the rule
is
applicable, i.e. for input {1} and interpretations 3 and 4 of the rule.
The desired result is {} if the rule is not applicable and the result denotes the

diseases of the patient.
The desired result is {1,2,3} if the rule is not applicable, and the result
denotes the
possible diseases of the patient.

Now let us compare this table with the results obtained from both approximate
reasoning models. Both models employ the formula

B'(y) = sup_x min (A'(x), J(A(x),B(y)))

In the Mamdani case, J is a t-norm, with the properties J(1,1)=1 and
J(0,1)=J(1,0)=J(0,0)=0.
In the second case (I call it the implication case), J is an implication
operator, with
the properties J(1,0)=0 and J(0,0)=J(0,1)=J(1,1)=1.

The result of the computation is:

Mamdani Implication
Input {1} | {1} {1}
Input {2} | {} {1,2,3}

So the Mamdani model produces the results which are desired in interpretation 3,
and the implication model produces the results which are desired in
interpretation 4.
One easily verifies that this holds true for other rules and inputs also.

So my conclusion for this example is : In case the interpretation of the rules is

interpretation 3, use the Mamdani model. In case the interpretation of the rules
is interpretation 4, use the implication model. In case the interpretation of the
rules
is interpretation 1 or interpretation 2, do not use any of these two models,
since
neither of them produces the desired results in case of crisp sets.

My conclusion in general is: In a practical situation, a simple test using crisp
sets can
determine which of the two models of approximate reasoning is appropriate, or
show
that neither of them is appropriate.

I look forward to receive your opinions!

greetings,

Pim van den Broek.

--------------63D9D56CE6B0D8F53C5FA784
Content-Type: text/html; charset=us-ascii
Content-Transfer-Encoding: 7bit

<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
Hello,
<p>A couple of weeks ago there was a discussion in this group following
a
<br>question of Andrej Albert on the difference between two models of
<br>approximate reasoning. Continuing this discussion I will consider a
small
<br>example, and show that the choice of the model should be dependent
on
<br>the interpretation of the rules, and moreover, that for some interpretations
<br>of the rules neither of the two models is appropriate.
<p>Suppose we have a set of rules of the form "IF A THEN B" where A is
a set
<br>of symptoms and B is a set of diseases. The intention of the rules
is to obtain
<br>a diagnosis of the disease(s) of a patient, given information about
the symptoms
<br>this patient exhibits. We will only consider crisp sets here, taking
the point of
<br>view that a model for approximate reasoning should produce reasonable
results
<br>also when only crisp sets are considered.
<br>Given a rule "IF A THEN B" where A is a set of symptoms and B is a
set of
<br>diseases and also given a set A' of symptoms, the resulting set B'
of diseases
<br>will depend on the meaning of the rules. A rule might be applicable
if the patient
<br>exhibits ANY of the symtoms, or if the patient exhibits ALL the symtoms.
<br>The result of the rule might mean that the patient suffers from ALL
diseases from
<br>the set, or that the disease is ANY of the diseases of the set.
<br>This leads to four possible interpretations of a rule "IF A THEN B":
<p>Interpretation 1 :&nbsp; IF the patient exhibits all symptoms from A
THEN the patient has
<br>all diseases of set B
<br>Interpretation 2 :&nbsp; IF the patient exhibits all symptoms from
A THEN the disease
<br>of this patient belongs to set B
<br>Interpretation 3 :&nbsp; IF the patient exhibits anyl symptoms from
A THEN the patient has
<br>all diseases of set B
<br>Interpretation 4 :&nbsp; IF the patient exhibits any symptoms from
A THEN the disease
<br>of this patient belongs to set B
<p>Now consider the case where there are 3 symtoms and 3 diseases, and
<br>symptoms and diseases both are denoted by the numbers 1,2 and 3.
<p>Consider the rule : IF {1,2} THEN {1}
<br>We consider two input sets of symptoms: {1} and {3}.
<br>The following table contains for both input sets and all four interpretations
of the
<br>rule the desired output set of diseases.
<p><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
Int. 1&nbsp;&nbsp; Int. 2&nbsp;&nbsp;&nbsp; Int. 3&nbsp;&nbsp; Int. 4</tt>
<br><tt>Input {1} |&nbsp; {}&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; {1,2,3}&nbsp;&nbsp;
{1}&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; {1}</tt>
<br><tt>Input {2} |&nbsp; {}&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; {1,2,3}&nbsp;&nbsp;
{}&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; {1,2,3}</tt>
<p>This table is computed as follows:
<br>The desired result is {1}&nbsp; ( the right hand-side of the rule),
in case the rule is
<br>applicable, i.e. for input {1} and interpretations 3 and 4 of the rule.
<br>The desired result is {} if the rule is not applicable and the result
denotes the
<br>diseases of the patient.
<br>The desired result is {1,2,3} if the rule is not applicable, and the
result denotes the
<br>possible diseases of the patient.
<p>Now let us compare this table with the results obtained from both approximate
<br>reasoning models. Both models employ the formula
<p><tt>B'(y) = sup_x min (A'(x), J(A(x),B(y)))</tt>
<p>In the Mamdani case, J is a t-norm, with the properties J(1,1)=1 and
<br>J(0,1)=J(1,0)=J(0,0)=0.
<br>In the second case (I call it the implication case), J is an implication
operator, with
<br>the properties J(1,0)=0 and J(0,0)=J(0,1)=J(1,1)=1.
<p>The result of the computation is:
<p><tt>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
Mamdani&nbsp;&nbsp; Implication</tt>
<br><tt>Input {1} |&nbsp; {1}&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; {1}</tt>
<br><tt>Input {2} |&nbsp; {}&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
{1,2,3}</tt>
<p>So the Mamdani model produces the results which are desired in interpretation
3,
<br>and the implication model produces the results which are desired in
interpretation 4.
<br>One easily verifies that this holds true for other rules and inputs
also.
<p>So my conclusion for this example is : In case the interpretation of
the rules is
<br>interpretation 3, use the Mamdani model. In case the interpretation
of the rules
<br>is interpretation 4, use the implication model. In case the interpretation
of the rules
<br>is interpretation 1 or interpretation 2, do not use any of these two
models, since
<br>neither of them produces the desired results in case of crisp sets.
<p>My conclusion in general is: In a practical situation, a simple test
using crisp sets can
<br>determine which of the two models of approximate reasoning is appropriate,
or show
<br>that neither of them is appropriate.
<p>I look forward to receive your opinions!
<p>greetings,
<p>Pim van den Broek.
<br>&nbsp;
<br>&nbsp;</html>

--------------63D9D56CE6B0D8F53C5FA784--

############################################################################
This message was posted through the fuzzy mailing list.
(1) To subscribe to this mailing list, send a message body of
"SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at
(2) To unsubscribe from this mailing list, send a message body of
"UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com"
to listproc@dbai.tuwien.ac.at
(3) To reach the human who maintains the list, send mail to
fuzzy-owner@dbai.tuwien.ac.at
(4) WWW access and other information on Fuzzy Sets and Logic see
http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info
(5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html

This archive was generated by hypermail 2b30 : Sun Dec 31 2000 - 04:40:52 MET