46838-s99 Machine Learning for Computational Finance Assignment 2

Due Monday March 29th 1999 at the beginning of class

Question 3.1 from the book
Question 3.2 from the book
Question 3.3 from the book
You are given the following decision tree for BuyStock for companies and a stock exchange you are unfamiliar with, along with the number of training examples which fell down each branch of the tree:
```
                 P:E Ratio  100 examples
                 /     \
           <7.5 /       \ >= 7.5
(10 examples)  /         \ (90 examples)
              /           \
           Assets        Assets
            / \           /  \
    > 2    /   \<=2   >16/    \ <= 16
   (2 ex.)/  (8 ex.) (18 ex.)  \ (72 ex.)
         /       \     /        \
       Yes       No   No       Yes

	
```
Now you have an example company for which you know the Assets (6) but you do not know the value for P:E Ratio. Can you use this decision tree to help you make your decision? How would you use it, if so? Why not, if not? If you know nothing about the company, can inspecting this tree give you any information about BuyStock ? Why? Why not?

ID3 searches for just one consistent hypothesis, whereas the CANDIDATE-ELIMINATION algorithm finds all consistent hyopotheses. Consider the correspondance between these two algorithms. Given the PayDividend example from Assignment 1, and the set of training examples (shown again in the table below):

Example	Price	Earnings	Sector	Market	Exchange	Outlook	Pay Dividend
1	Up	High	Manufacturing	Bull	NYSE	Strong	Yes
2	Up	High	Service	Bull	NYSE	Strong	Yes
3	Down	Low	Service	Bull	NYSE	Weak	No
4	Up	High	Service	Bull	Nasdaq	Weak	Yes

(a)Show the decision tree that would be learned by ID3 given these training examples.
(b)What is the relationship between the decision tree and the version space that is learned from these same examples? Is the decision tree equivalent to one of the members of the version space?

Read the paper distributed in class Inducing Stock Screening Rules for Portfolio Induction. Note the assumptions referred to in the conclusion: (1) the set of variables chosen are appropriate for class description, (2) members of each class exhibit regularities than can be expressed by conditions imposed on the chosen variables.
- (a) What will ID3 do when members of the class do not exhibit regularities that can be expressed by conditions on the chosen variables?
- (b) Suppose a decision tree generates an unintuitive rule over the data. What might you choose to do about this rule?
- (c) How could you choose variables to provide as input to ID3? How could you tell if they exhibited interesting regularities?

Rosie Jones

Last modified: Mon Mar 15 13:45:01 EST 1999