Review1

• What is NFL? What are its implications for selecting the "best" learner
• List 5 baseline criteria for an AI tools, make argument:
  • Why it is important to meet that criteria?
  • Why, pragmatically, it might be necessary to ignore (or, at least relax) that criteria
• Distinguish supervised (S) from unsupervised (U) learning
  • When would do S or U?
  • Is FFT an S or U? Why? (Hint: your answer should say something about FFT).
  • Is the Fastmap Cluster an S or U? Why? (Hint: your answer should say something about the Fastmap clusterer).
• Instance models refer to specific values (e.g. X=6,Y=10,Z=100) while other models refer to ranges of some attributes (e.g. X>6 and Y < 5).
  • Which of these models reference points or volumes?
  • Which of these offer more generalizations of the past?
  • Which of these are shorter to share?
  • How to generalize from point models to models that cover volumes?
• FFTs:
  • What are the attributes in an node of an FFT tree?
  • What is the structure of an FFT tree (hint: use the node attributes to make that description).
• Fastmap Clustering
  • Given 2 points Y,Z at distance c, if a new point X is a=dist(X,Y) and b=dist(X,Z), derive an expression for the distance x that X falls along the line between Y,Z.
  • What are the attributes in an node of an Fastmap cluster? (hint: parts of it are recursive)
  • What is the structure of an Fastmap clustering tree (hint: use the node attributes to make that description).
  • How to use a Fastmap cluster node for
    • classification (hint: there are many ways)
  • anomaly detection, sharing, privacy (hint: see III.D.4
    • incremental model updates over an infinite stream of data? (hint
• How to use any clustering method for optimization, anomaly detection, classification, sharing, privacy.
• How to use a Fastmap cluster free for: - very fast optimization (hint: see Algorithm1)
• What are diversity measures for numeric and symbolic values? Offer a formula for each.
• What is the diversity of "y,n,y,y,y,n,y,y,n"? If you don't have a calculator, show all working and stop before the final calculation.
• What is the diversity of "10,89,32,11,9,90,30,31,91"? If you don't have a calculator, show all working and stop before the final calculation.
• Consider the following data.
  • Just considering the age column, where to divide it such that age diversity is minimized? Assume a minimum bin size of 3.
  • Now consider the age,alive relationship. Where to divide age in order to reduce the diversity of alive? Assume a minimum bin size of 3.

Data:

 age,alive
 10, y
 89, n
 32, y
 11, y
 9,  y
 90, n
 30, y
 31, y
 91, n

• Iterative dichmotization (ID) algorithms divide attributes into ranges, then recurse on each range.
  • How do ID algorithms decide what attribute with which to split the data