Re: Can a fuzzy model be made from an incomplete set of variables?

From: WSiler@aol.com
Date: Thu Aug 23 2001 - 07:16:36 MET DST

  • Next message: Andrzej Pownuk: "Re: Thomas' Fuzziness and Probability"

    In a message dated 8/22/2001 1:23:26 PM Eastern Daylight Time,
    aquila.em.hx.deere.com@hx.deere.com writes:

    > , I am trying to use NeuroFuzzy technique to map a number of input =
    > variables to an output variable. The data are from field measurements = and
    > are very vague. In addition, some of the input variables are not =
    > measurable although they are the less important variables. I am wondering
    >

    The answer is Yes, but not with the conventional fuzzy rules. You need to use
    the method of William Combs. Briefly, the method is this.

    Suppose you have two input variables, fuzzified into two discrete fuzzy sets,
    say Size with members Small and Big, and Speed with members Slow and Fast,
    and one output discrete fuzzy set with members Little and Big..

    Ordinarily you would write four rules, like this:

    IF (x1 is Small and x2 is Slow) THEN y is Big
    IF (x1 is Big and x2 is Slow) THEN y is Little
    IF (x1 is Small and x2 is Fast) THEN y is Big
    IF (x1 is Big and x2 is Fast) THEN y is Little

    This is what Combs calls the Intersection Rule Configuration, or IRC. There
    are two problems with this. First, the number of rules grows exponentially
    with the number of input variables. (With 20 input variables, you will have a
    huge number of rules.) Second, if say x2 is missing you are dead; no rules
    will fire, and you will get no answer.

    Combs proposes the Union Rule Configuration, or URC. In the Combs method,
    yhou rules would be:

    IF (x1 is Small) THEN y is Big
    IF (x2 is Slow) THEN y is Little
    IF (x1 is Small) THEN y is Big
    IF (x2 is Fast) THEN y is Little
    IF (x1 is Small) THEN y is Big
    IF (x2 is Slow) THEN y is Little
    IF (x1 is Small) THEN y is Big
    IF (x2 is Fast) THEN y is Little

    As each rule fires, we will get different values for the grade of membership
    of Big or Little, say Big1, Big2, Big3, Big4 and Little1, Little2, Little3
    and Little4. These are aggregated by simply taking the average value: the
    final value for Big's greade of membership would be

    Big = (Big1+Big2+Big3+Big4)/4, and
    Little = (Little1+Little2+Little3+Little4)/4.

    Although in this case the Combs URC method has twice the number or rule as
    the IRC, the number of rules increases LINEARLY with the number of input
    variables, so handling 20 input variables still produces a manageable number
    of rules. Further, missing data for some input variables will not stop you
    from getting an answer, although the answer might not be quite as reliable as
    if no input data were missing.

    I have checked the method out on a number of problems, and it works, although
    the answers from the URC and IRC can be slightly different.

    William Siler

    --part1_77.19da2afe.28b5b575_boundary
    Content-Type: text/html; charset="US-ASCII"
    Content-Transfer-Encoding: 7bit

    <HTML><FONT FACE=arial,helvetica><FONT SIZE=2>In a message dated 8/22/2001 1:23:26 PM Eastern Daylight Time,
    <BR>aquila.em.hx.deere.com@hx.deere.com writes:
    <BR>
    <BR>
    <BR><BLOCKQUOTE TYPE=CITE style="BORDER-LEFT: #0000ff 2px solid; MARGIN-LEFT: 5px; MARGIN-RIGHT: 0px; PADDING-LEFT: 5px">, I am trying to use NeuroFuzzy technique to map a number of input =
    <BR>variables to an output variable. The data are from field measurements = and
    <BR>are very vague. In addition, some of the input variables are not =
    <BR>measurable although they are the less important variables. I am wondering
    <BR>if a fuzzy model can be made from an incomplete set of = variables.=20 </BLOCKQUOTE>
    <BR>
    <BR>The answer is Yes, but not with the conventional fuzzy rules. You need to use
    <BR>the method of &nbsp;William Combs. Briefly, the method is this.
    <BR>
    <BR>Suppose you have two input variables, fuzzified into two discrete fuzzy sets,
    <BR>say Size with members Small and Big, and Speed with members Slow and Fast,
    <BR>and one output discrete fuzzy set with members Little and Big..
    <BR>
    <BR>Ordinarily you would write four rules, like this:
    <BR>
    <BR>IF (x1 is Small and x2 is Slow) THEN y is Big
    <BR>IF (x1 is Big and x2 is Slow) THEN y is Little
    <BR>IF (x1 is Small and x2 is Fast) THEN y is Big
    <BR>IF (x1 is Big and x2 is Fast) THEN y is Little
    <BR>
    <BR>This is what Combs calls the Intersection Rule Configuration, or IRC. There
    <BR>are two problems with this. First, the number of rules grows exponentially
    <BR>with the number of input variables. (With 20 input variables, you will have a
    <BR>huge number of rules.) Second, if say x2 is missing you are dead; no rules
    <BR>will fire, and you will get no answer.
    <BR>
    <BR>Combs proposes the Union Rule Configuration, or URC. In the Combs method,
    <BR>yhou rules would be:
    <BR>
    <BR>IF (x1 is Small) THEN y is Big
    <BR>IF (x2 is Slow) THEN y is Little
    <BR>IF (x1 is Small) THEN y is Big
    <BR>IF (x2 is Fast) THEN y is Little
    <BR>IF (x1 is Small) THEN y is Big
    <BR>IF (x2 is Slow) THEN y is Little
    <BR>IF (x1 is Small) THEN y is Big
    <BR>IF (x2 is Fast) THEN y is Little
    <BR>
    <BR>As each rule fires, we will get different values for the grade of membership
    <BR>of Big or Little, say Big1, Big2, Big3, Big4 and Little1, Little2, Little3
    <BR>and Little4. These are aggregated by simply taking the average value: the
    <BR>final value for Big's greade of membership would be
    <BR>
    <BR>Big = (Big1+Big2+Big3+Big4)/4, and
    <BR>Little = (Little1+Little2+Little3+Little4)/4.
    <BR>
    <BR>Although in this case the Combs URC method has twice the number or rule as
    <BR>the IRC, the number of rules increases LINEARLY with the number of input
    <BR>variables, so handling 20 input variables still produces a manageable number
    <BR>of rules. Further, missing data for some input variables will not stop you
    <BR>from getting an answer, although the answer might not be quite as reliable as
    <BR>if no input data were missing.
    <BR>
    <BR>I have checked the method out on a number of problems, and it works, although
    <BR>the answers from the URC and IRC can be slightly different.
    <BR>
    <BR>William Siler
    <BR></FONT></HTML>

    --part1_77.19da2afe.28b5b575_boundary--

    ############################################################################
    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 : Thu Aug 23 2001 - 07:22:01 MET DST