18.04.2013 Views

The.Algorithm.Design.Manual.Springer-Verlag.1998

The.Algorithm.Design.Manual.Springer-Verlag.1998

The.Algorithm.Design.Manual.Springer-Verlag.1998

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

War Story: Nothing but Nets<br />

``You are right. We should win if we can break big nets into small nets. We want the nets to be small,<br />

both in the number of points and in the area of the net. But we must be sure that if we validate the<br />

connectivity of each small net, we will have confirmed that the big net is connected. Whenever there is<br />

one point in common between two little nets, that is enough to show that the bigger net formed by the<br />

two little nets is connected, since current can flow between any pair of points.''<br />

Now we had to break each net into overlapping pieces, where each piece was small. This is a clustering<br />

problem. Minimum spanning trees are often used for clustering, as discussed in Section . In fact, that<br />

was the answer! We could find the minimum spanning tree of the net points and break it into little<br />

clusters whenever a spanning tree edge got too long. As shown in Figure (d), each cluster would share<br />

exactly one point in common with another cluster, with connectivity ensured because we are covering the<br />

edges of a spanning tree. <strong>The</strong> shape of the clusters would reflect the points in the net, exactly as we<br />

would want. If the points lay along a line across the board, the minimum spanning tree would be a path,<br />

and the clusters would be pairs of points. If the points all fell in a tight region, there would be one nice fat<br />

cluster that the right arm would just scoot around.<br />

So I explained the idea of constructing the minimum spanning tree of a graph. <strong>The</strong> boss listened,<br />

scribbled more notes, and frowned again.<br />

``I like your clustering idea. But these minimum spanning trees you talk about are defined on graphs. All<br />

you got are points. Where do the weights of the edges come from?''<br />

``Oh, we can think of it as a complete graph, where every pair of points are connected. <strong>The</strong> weight of the<br />

edge defined by the two points is simply the distance. Or is it...?''<br />

I went back to thinking. <strong>The</strong> edge cost between two points should reflect the travel time between them.<br />

While the distance was related to the travel time, it wasn't necessarily exactly the same thing.<br />

``Hey. I have a question about your robot. Does it take the same amount of time to move the arm leftright<br />

as it does up-down?''<br />

<strong>The</strong>y thought a minute. ``Yeah, it does. We use the same type of motors to control horizontal and vertical<br />

movements. Further, since the two motors for each arm are independent, we can simultaneously move<br />

each arm both horizontally and vertically.''<br />

``That so? <strong>The</strong> time to move both one foot left and one foot up is exactly the same as just moving one<br />

foot left? This means that the weight cost for each edge in the graph should not be the Euclidean distance<br />

between the two points, but the biggest difference between either the x- or y-coordinate. This is<br />

something we call the metric, but we can capture it by changing the edge weights in the graph.<br />

Anything else funny about your robots?'' I asked.<br />

file:///E|/BOOK/BOOK2/NODE79.HTM (3 of 4) [19/1/2003 1:29:19]

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

Saved successfully!

Ooh no, something went wrong!