Date: Tue, 16 Mar 93 11:38:43 -0700
From: kevin@axon.cs.byu.edu (Kevin Vanhorn)
To: rick@axon, george@axon, cory@axon, martinez@bunsen.cs.byu.edu,
        vanhorn@bert.cs.byu.edu, cgc@axon, tim@axon, randy@axon, dan@axon
Subject: Random ML problem generator


In the directory .../mldb/rand I have installed the following programs
for generating random machine learning problems:

        genex1_d, genex1_h, genex2_d, genex2_h.

The "_d" programs run on the DECstations, and the "_h" programs run on
the HP's.  The argument lists are as follows:

  genex1 <rules> <classes> <inputs> <ferr> <traini> <testi> <trn> <tst>
    WHERE
    <rules> is the number of rules in the randomly-generated ordered set
      of production rules which is the target hypothesis
    <classes> is the number of possible different classes
    <inputs> is the number of possible binary inputs
    <ferr> is the fraction of noisy examples
    <traini> is the number of examples to generate for the training set
    <testi> is the number of examples to generate for the testing set
    <trn> is the name of the file to which the training examples are written
    <tst> is the name of the file to whcih the test examples are written.

The files <trn> and <tst> are written in the "num" HUT format.  Genex1 uses
a uniform distribution over the input space, and chooses the number of
literals in each rule's test so as to ensure that each rule (including the
default) covers an approximately equal fraction of the input space.

  genex2 <size> <rules> <classes> <inputs> <ferr> <traini> <testi> <trn> <tst>
    WHERE
      <size> is the total number of literals used in the target hypothesis
      the other arguments are the same as for genex1.

Genex2 does *not* use a uniform distribution over the input space.  Instead,
there is an equal probability that an input will be covered by any of the
<rules>+1 rules plus default, and a uniform distribution is used within the
set of examples covered by any rule (including default).

(Note: When I say that a rule "covers" an input vector I mean
that it is the first rule that matches the input vector.)

Please report to me any problems in using these programs.

-----------------------------------------------------------------------------
Kevin S. Van Horn       | It is the means that determine the ends.
vanhorn@bert.cs.byu.edu |