13.07.2015 Views

Data Mining: Practical Machine Learning Tools and ... - LIDeCC

Data Mining: Practical Machine Learning Tools and ... - LIDeCC

Data Mining: Practical Machine Learning Tools and ... - LIDeCC

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.

74 CHAPTER 3 | OUTPUT: KNOWLEDGE REPRESENTATIONTable 3.2Training data for the shapes problem.Width Height Sides Class2 4 4 st<strong>and</strong>ing3 6 4 st<strong>and</strong>ing4 3 4 lying7 8 3 st<strong>and</strong>ing7 6 3 lying2 9 4 st<strong>and</strong>ing9 1 4 lying10 2 3 lyingmation the learning algorithm will be given is the width, height, <strong>and</strong> number ofsides of each block. The training data is shown in Table 3.2.A propositional rule set that might be produced for this data is:if width ≥ 3.5 <strong>and</strong> height < 7.0 then lyingif height ≥ 3.5 then st<strong>and</strong>ingIn case you’re wondering, 3.5 is chosen as the breakpoint for width because it ishalfway between the width of the thinnest lying block, namely 4, <strong>and</strong> the widthof the fattest st<strong>and</strong>ing block whose height is less than 7, namely 3. Also, 7.0 ischosen as the breakpoint for height because it is halfway between the height ofthe tallest lying block, namely 6, <strong>and</strong> the shortest st<strong>and</strong>ing block whose widthis greater than 3.5, namely 8. It is common to place numeric thresholds halfwaybetween the values that delimit the boundaries of a concept.Although these two rules work well on the examples given, they are not verygood. Many new blocks would not be classified by either rule (e.g., one withwidth 1 <strong>and</strong> height 2), <strong>and</strong> it is easy to devise many legitimate blocks that therules would not fit.A person classifying the eight blocks would probably notice that“st<strong>and</strong>ing blocks are those that are taller than they are wide.” This rule doesnot compare attribute values with constants, it compares attributes with eachother:if width > height then lyingif height > width then st<strong>and</strong>ingThe actual values of the height <strong>and</strong> width attributes are not important; just theresult of comparing the two. Rules of this form are called relational, becausethey express relationships between attributes, rather than propositional, whichdenotes a fact about just one attribute.

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

Saved successfully!

Ooh no, something went wrong!