Computer Science 315
Artificial Intelligence

Assignment 5 - AIMIA Chapter 18: Learning from Observations

Did you ever have to finally decide
Say yes to one and let the other one ride
There's so many changes and tears you must hide
Did you ever have to finally decide?

– The Lovin' Spoonful, "Did You Ever Have to Make Up Your Mind" (1965)

Due Friday, 09 November

Implement the DECISION-TREE-LEARNING algorithm (Figure 18.5, page 658) in Python, as a class named DTL contained in a file dtl.py. Here's the DTL API. The following code will be useful in writing DTL: The following code will help you test your DTL: You should write dtl.py following the interface provided. Note that you must also implement the constructor for this class, as well as a test function, to make decisions on novel examples. Your restaurant decision tree should give the right answer for the novel examples. Initially, you should just write your CHOOSE-ATTRIBUTE function to return the first attribute in attributes. Then you can modify the function to use information theory, as described on pages 659-660 (Choosing Attribute Tests). In particular, you should choose the attribute with the maximum gain. My code (andtest.py, restaurant.py) will test your code with some examples, but you should also do a "reality check": for each example in the table on page 656, make sure that your tree makes the right decision. If your test function breaks on the examples I've set up, there is not necessarily a bug in your code; the problem could be caused by numerical precision differences on different systems. If this happens, and you're convinced that your learning function works, edit the last lines of Restaurant.py to reflect the kind of tree you built, and send me Restaurant.py along with dtl.py. Of course, you should check that your tree makes sense!