CS322 Fall 1997
Assignment 11
Due: 1:30pm, Friday 28 November 1997.
In the file ~cs322/cilog/dtlearn1.pl is an axiomatization of a
decision-tree learning algorithm. This works with the data in the file
~cs322/cilog/dtlearn_t1.pl, and produces the tree of Figure
11.3 of the textbook.
This uses a complicated procedure to determine what is the best
attribute to split on. In this question, you will look at what happens
when this procedure is changed.
-
Suppose you change the procedure
select_split so that it always
selects the first element of the list of attributes. What tree is found when
the attributes are in the order [author, thread, length,
where_read]? Does this tree represent a different function than
that found with the more complicated selection mechanism? Explain.
-
What tree is found when
the attributes are in the order [where_read, thread,
length, author]? Does this tree represent a different function than
that found with the more complicated selection mechanism or the one
given for the preceding part? Explain.
-
Is there a tree that correctly classifies the training examples, but
represents a different function than those found by the preceding
algorithms?
If so, give it. If not, explain why.
Suppose that in the output of a neural network, we assign any value
over 0.5 to be true and any less than 0.5 to be false (i.e., any
positive value before the activation function is true, and a negative
value is false).
Given the final parameter settings, of Figure 11.7 of the text, give a
logical formula (or a decision tree or a set of rules) that represents
the Boolean function that is the value for the hidden units and the
output units. This formula or set of rules shouldn't refer to any real
numbers.
Hint: one brute-force method is to go through the 16 combinations of
values for the inputs to each hidden unit and determine the truth
value of the output. A better method is to try to understand the
functions themselves. Your meta-interpreter for parameterized logic
programs may help here.
For each question in this assignment: how long you spent on it, was it
reasonable and what did you learn?