Skip to content

Latest commit

 

History

History
88 lines (75 loc) · 3.7 KB

review1.md

File metadata and controls

88 lines (75 loc) · 3.7 KB

 

home | copyright ©2019, tjmenzie@ncsu.edu

syllabus | src | submit | chat

Review1

For Numbers:

  • Define mean and standard deviation.
  • What does a large sd mean?
  • What does a small sd mean?
  • Show a list of numbers for which the standard deviation is zero

For Symbols:

  • There are two terms term1,term2 that are analogous to mean and standard deviation. What are they?
  • Given the set of symbols "aaaabbc", calculate those terms (it is ok to leave the result as fractions).
  • Adjust the set of symbols such that term1 changes value.
  • Adjust the set of symbols such that term2 increases or decreases.
  • Show a list of symbols for which the "term2" is zero.

For each of TOOL ∈ (blusterers, classifiers, regressors, and multi-objective optimizers).

  • Define what TOOL does

  • Give a specific example of when you would use TOOL.

  • (Here are two questions you may not be able to answer... yet)

    • Name a commonly used algorithm for doing TOOL;
    • Very briefly, describe how TOOL operators.

  • Assume data headers can have the special characters;

    • "!" (for "class")
    • "<" for ("minimize"),
    • ">" for (maximize),
    • "$" (for number)
    • and that anything without any mark is an independent symbol.
    • Assume a data set has five columns with a header name,age,daysTillDeath, zipCode,income
      • Write down a header that means we should use TOOL.

  • Assuming A,B,C,D are the true negative, false negatives, false positives, true positives (respectively) seen by a classifier , then

    • define accuracy
    • define recall
    • define false alarm
    • define precision
    • When is "accuracy not accurate"? Give specific values to A,B,C,D where a classifier has a high accuracy yet usually misses the target concept (i.e. high accuracy, low recall)
    • When is "precision not precise"? i.e. Give a specific example where a classifier has a very low precision, yet still might be considered useful.
    • The following question using the following function.
      • What is the accuracy seen below?
      • What is the recall for "yes"?
      • What is the false alarm rate for "no"?
      • What is the precision for "maybe"?
function _abcd(f,i,j) {
  Abcd(i)
                             # want,    got
                             #------    -------
  for(j=1; j<=6; j++) Abcd1(i,"yes",    "yes")
  for(j=1; j<=2; j++) Abcd1(i,"no",     "no")
  for(j=1; j<=5; j++) Abcd1(i,"maybe",  "maybe")
  for(j=1; i<=1; j++) Abcd1(i,"maybe",  "no")
  AbcdReport(i)
}

Multi-objective optimization

  • define the Boolean domination predicate suitable for two objectives
  • Assuming we want to minimize power consumption and cost
    • draw 20 dots on a two-d grid. Mark the Pareto frontier.
      • on that first drawn, draw a tiny square around any dot "X" then draw the (much larger) rectangle indicating which other dots are dominated by "X"
    • Make a second drawing (on a new pierce of paper)
      • draw 10 circles from optimizer one and 10 crosses from optimizer two
      • draw the reference frontier
  • Define hypervolume, spread, generational distance (GD), inverse generational distance (IGD)
    • On the second drawing, show one distance measurement that would be make for GD but not for IGD
  • (Here are two questions you may not be able to answer... yet)
    • define a domination predicate suitable for 3,4,5 objectives.