Marking guide for COMP8400 assignment 2, 2009
=============================================

General comments: Please follow exactly the questions asked in
  assignment specifications, as otherwise it is very hard to do an
  accurate marking!

***** Part 1: 6 marks *****

      1 mark: Correct data set generated.
      1 mark: Correct and complete length 2 frequent item-sets.
      1 mark: Correct and complete length 3 frequent item-sets.
      1 mark: Correct rules generated and their correct support
              values.
      1 mark: Correct confidence values for these rules.
      1 mark: Correct lift values for these rules.

      -0.5 mark for small errors like missing item-sets, wrong counts,
           etc.

----------------------------------------------------------------------

***** Part 2: 3 marks *****

      1 mark: Correct confusion matrix, including 'Total' number and
              normalised confusion matrix (percentages).
      1 mark: Correct accuracy and error rate percentage numbers.
      1 mark: Correct specificity, precision and recall percentage
              numbers.

      No penalties for small rounding errors in parts 2 and 3.

----------------------------------------------------------------------

***** Part 3 A: 6 marks *****

      1 mark: For executive summary, if 10 lines of text or less, and
              if it addresses the classification project conducted,
              including data sets summary and classification outcomes.
        -0.5 if the name or size (number of records) of the data set
              is not given.

      0.5 mark: Data set description, including name, source, number
                of records, attributes in data, description of
                attributes, etc.

      1 mark: Data exploration steps, data quality description,
              including missing values, distribution of values, plots,
              etc. (a correlation plot would have been nice). Also,
              if only a subset of attributes are shown in plots you
              need to explain why you have chosen these attributes and
              not others.

      1 mark: Data cleaning and transformation steps performed,
              including feature construction if necessary.
              Converting numerical into categorical variables was not
              necessary, but considering how to handle records with
              missing values was important.
              Normalisation is (usually) not necessary for
              classification, and neither is categorisation of
              numerical variables into categoricals.
              Also, start a classification by using all (or most)
              variables (unless some are very correlated with each
              other), and don't manually select a sub-set (unless you
              have a very large number of variables).

      1 mark: Description of classification approaches, including
              classifiers chosen, their parameters, etc.
              One paragraph per classifier. Here you needed to explain
              you chose parameter settings. You lost marks if you did
              not explain why a certain parameter value/setting was
              used.

      1 mark: Presentation of classifier results, should include
              confusion matrix (both raw numbers and percentages),
              and visualisation of the results, e.g. ROC,
              precision-recall curves, etc. The best would have been
              to show one ROC plot containing all three classifier
              results (easier to compare than three separate plots).
              (0.5 mark for confusion matrix, 0.5 for graphs)

      0.5 mark: General description of your classification project,
                including problems, things learned, etc.

----------------------------------------------------------------------

***** Part 3 A: 6 marks *****

    * two marks (2) for the program listing, its readability,
      structure, commenting, etc.
      -0.5 if the Canberra distance function did not check for
           division by zero in the formula!
      -0.25 for bad variable / method / function naming.
      -0.25 for lost of code in the main function / method and not in
            separate functions / methods.
      - 0.25 for missing or unclear comments.

    * half a mark (0.5) for your description of how your program
      works.

    * half a mark (0.5) for your description of how you tested your
      program.
      -0.25 if testing not properly explain, or only limited testing
            described.

    * one mark (1) each for correctly classifying all the test records
      in the three test input files.