COSC 159
                Assignment #3

         Due date: Wednesday, March 11

1. In class we saw how decision trees can be used to record and
generalize training examples by looking at the book's example
of recycling bins in offices.  In this problem, we'll consider
the following training examples for a person's book preferences:

  Title      |  Binding   Reality    Length    Topic     | Like?
-------------+-------------------------------------------+-----
Moby Dick    | Hardcover  Fiction     Long    Adventure  | YES
             |                                           |
War & Peace  | Paperback  Fiction     Long    Politics   | YES
             |                                           |
Hawaii       | Hardcover  Fiction     Long    History    | YES
             |                                           |
Perestroika  | Hardcover  Nonfiction  Medium  Politics   | NO
             |                                           |
Star Trek    | Paperback  Fiction     Medium  Science    | NO
             |                                           |
Bugs         | Paperback  Nonfiction  Short   Science    | YES
             |                                           |
Flatland     | Hardcover  Fiction     Short   Adventure  | NO
             |                                           |
Dune         | Paperback  Fiction     Long    Science    | YES
             |                                           |
Bolivia      | Paperback  Nonfiction  Medium  History    | NO

For each of the four attributes (Binding, Reality, Length, Topic),
compute the class partitions and disorder number.  In other words,
produce a table like Table 5.2 on page 201 of the text.  Using this
disorder information, say which attribute should be chosen for the
root (top) node of our decision tree.
  While you should be able to complete this assignment by hand with
a calculator, it is fine with me if you wish to use some or all
of the decision tree code provided at the end of Chapter 5 (and
in ~mikes/159/CodeFromText/05.Learning on studsys).  Just explain
how you did the computations and give me a listing of any code
that you write or modify.

2. Using the Lisp code provided in ~mikes/159/CodeFromText/05.Learning
on studsys, run some examples of neural nets learning the "equivalent"
logical connective.  In particular, I'd like you to run the networks
defined in pdp1.lisp and pdp2.lisp (which match those defined in
the textbook on pages 214 and 219) and use the function average 
(which is defined in the file pdp.lisp and is automatically loaded) to
collect statistics on how quickly each learns the "equivalent" concept.
You will need to define training data for the "equivalent" connective.
This logical function takes two inputs and returns TRUE if both
inputs are TRUE or if both are FALSE; it returns FALSE if one input
is TRUE and the other FALSE.  You should report an average and a
success rate for each net based on at least 10 attempts (the number
of attempts is one of the parameters to the function average).
A log of some net experiments can be found in the
file ~mikes/159/annlog on studsys.