====== Learning Decision Trees ======

Material:
  * {{:public:t-622-arti-13-1:dtl_learning.zip|java project}}
  * [[http://aima-java.googlecode.com/svn/trunk/aima-all/release/javadoc/aima-core/index.html|javadoc for the framework]]

The zip archive contains three data sets and Java code to learn a decision tree from those data sets. Your task is to check how well the learning algorithm performs on the given data sets.

Tasks:
  - Change the existing code, so that it prints the necessary data for a learning curve. That is, change the code such that it learns trees for increasing numbers of training examples and test each of the trees using the test set.
  - Plot the learning curves for all three data sets (e.g., using a spreadsheet application of your choice) (monk-1, monk-2, monk-3) and interpret them.
  - Are all the trees that are learned consistent with the training data (consistent = all examples are classified correctly)? If not, what could be the reason? (Hint: Consistency of the tree with the training set can be checked easily by adding a single line of code.)
  - Look at the true functions for the three data sets in monk.names (Section 9). Design a good decision tree for the concept of monks-1 by hand.
  - Compare this decision trees to the one that was learned by the algorithm using the whole training set.


Hand In:
  - Learning curves for the three data sets.
  - Interpretation of the learning curves.
  - Answer to 3.
  - Decision tree for 4.
  - Interpretation of your findings for 5.