Summary

• Decision Trees are practical for discrete-valued functions, grows tree from root down, selecting next best attribute at each new node added to tree.
• ID3 searches complete hypothesis space. It can represent any discrete-valued function defined over discrete valued instances, therefore it avoids the problem of the target function not being in the hypothesis space.
• Inductive Bias implicit in ID3 is for smaller trees, only grows as large as needed to classify training examples