18.11.2012 Views

anytime algorithms for learning anytime classifiers saher ... - Technion

anytime algorithms for learning anytime classifiers saher ... - Technion

anytime algorithms for learning anytime classifiers saher ... - Technion

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

<strong>Technion</strong> - Computer Science Department - Ph.D. Thesis PHD-2008-12 - 2008<br />

Procedure IIDT(E, A)<br />

T ← Greedy-Build-Tree(E, A)<br />

While not-interrupted<br />

node ← Choose-Node(T, E, A)<br />

t ← subtree of T rooted at node<br />

r ← Next-R(node)<br />

Enode ← Examples-At(node, T, E)<br />

Anode ← Attributes-At(node, T, A)<br />

t ′ ← Rebuild-Tree(Enode, Anode, r)<br />

If Better(t ′ , t)<br />

replace t with t ′<br />

Return T<br />

Procedure Examples-At(node, T, E)<br />

Return {e ∈ E | e reaches node}<br />

Procedure Attributes-At(node, T, A)<br />

Return {a ∈ A | a /∈ ancestor of node}∪<br />

{a ∈ A | a is numeric}<br />

Procedure Next-R(node)<br />

If Last-r(node) = 0<br />

Return 1<br />

Else<br />

Return 2 · Last-R(node)<br />

Figure 3.12: IIDT: framework <strong>for</strong> interruptible induction of decision trees<br />

candidate <strong>for</strong> such an algorithm is LSID3, which is expected to produce better<br />

subtrees when invoked with a higher resource allocation. In what follows we focus<br />

on the different components of IIDT and suggest a possible implementation that<br />

uses LSID3 <strong>for</strong> revising subtrees.<br />

Choosing a Subtree to Improve<br />

Intuitively, the next node we would like to improve is the one with the highest<br />

expected marginal utility, i.e., the one with the highest ratio between the expected<br />

benefit and the expected cost (Horvitz, 1990; Russell & Wefald, 1989). Estimating<br />

the expected gain and expected cost of rebuilding a subtree is a difficult problem.<br />

There is no apparent way to estimate the expected improvement in terms of either<br />

37

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!