CMPT 459 (00-3): Assignment 2

(100 marks)

Due time and date: 13:30 on December 1, 2000, Friday.

Please hand in your assignment to the instructor.

Submission Requirements

• All answers must be computer generated (including text and diagrams).
• The hand-in version must be ordered correctly and stapled in the top left corner.
• The hand-in version must include a header page indicating: student name, student number, user id, course number and assignment number.

Part 1: Exercise Assignment (60 points)

• Question 1 (14 points): Exercise 6.13 on page 274 in the text book.
• Question 2 (16 points): Exercise 6.17 on page 275 in the text book.
• Question 3 (15 points): Exercise 7.6 on page 329 in the text book.
• Question 4 (15 points): Exercise 8.12 on page 223 in the text book.

Part 2: Program Assignment (40 points)

1. Using DBMiner to Mine Association Rules (20 Marks)

• Required Setup:

  (1) Follow the menu route CMPT459->DBMiner to start up DBMiner.

  (2) When DBMiner is ready to use, check the on-line DBMiner tutorial which can be accessed from your course home page and also from here.

  (3) Following the steps outlined in Section 1 in the tutorial to connect DBMiner to the OLAP server, the name of which is CYPRESS in the CSIL.

  (4) Read Section 2 in the tutorial on how to mine associations from a data cube in DBMiner. The tutorial shows how to mine inter-dimensional associations. In this question, you are required to mine both inter-dimensional and intra-dimentional asscociations. Pay special attention to step (d) in the tutorial in that section. For intra-dimensional association mining in DBMiner, you will specify the Repetitive dimension as well as the Group by dimension(s) in that step. You may also set the contraints there (you may need to read the on-line documentation on what the contraints are in DBMiner).

  (5) In this question, select FoodMart as your data base.

   

• Tasks:

   

  • Step (1) Suppose you have an inter-dimensional association query as follows.

          MINE Inter-dimensional association
          WITH RESPECT TO Customers, Education Level, Gender, Marital Status, Product
          FROM CUBE FoodMart-Sales
          SET Minimum support 5%
          SET Minimum confidence 45%
          WITH CONSTRAINTS AS FOLLOWS
               The level for dimension Customers: USA.
               The level for dimension Education Level: Bachelor degree
    

    Use the mining wizard in DBMiner to submit this association task. How many rules did you obtain finally? For each meta rule below, report the one with the highest support in the final rule set.

     

    1. W and X => Y and Z
    2. X => Y and Z
    3. X => Y

    where W, X, Y, and Z are different attribute-value pairs.

  • step (2) For the above query, keep the same specifications you set in step (1). But for this step, dynamically change the Minimum support back and forth until the number of rules in the final rule set falls between 20-25.

    How many rules did you obtain finally? For each meta rule below, report the one with the highest support in the final rule set.

     

    1. X => Y
    2. X => Y and Z

    where X, Y and Z are different attribute-value pairs.

    In the final rule set, do you think whether there are any redundant rules? If yes, choose one and explain why it is redundant.

  • Step (3) Suppose you have an intra-dimensional association query as follows.

          MINE Intra-dimensional association
          WITH REPETITIVE DIMENSION Promotion Media
          GROUP BY DIMENSION Customers
          FROM CUBE FoodMart-Sales
          SET Minimum support 10%
          SET Minimum confidence 80%
          WITH CONSTRAINTS AS FOLLOWS
               The level for dimension Customers: [City]
               The items selected for dimension Promotion Media to do the association: 
                   [Bulk Media], [Cash Register Handout], [Daily Paper, Radio, TV],
                   [Daily Paper, Radio], [Daily Paper], [In-Store Coupon]
               Only use these constraints to do the association
    

    Use the mining wizard in DBMiner to submit this association task. Please note that when you submit this query, follow the constraints strictly and do not change them. Otherwise, it would generate too many rules and cost too much time.

    How many rules did you finally obtain? List the three rules with the highest support along with their corresponding confidence.

  • Step (4) Use the same query as in Step (3). Dynamically change the Minimum support until there is at least one rule but no more than three rules left in the final rule set. Show what Minimum support you chose at this time. List these rules along with their support as well as their confidence.
• Submission:

  For each step, submit the result you obtained.

2. Using DBMiner to Mine Classification Rules (20 Marks)

• Required Setup:

  (1) Use the same setup procedures outlined in Question 3 to start up DBMiner and connect it to the OLAP server, CYPRESS. Read Section 3 in the tutorial on how to mine classifications from a data cube in DBMiner.

  (2) Read the on-line documentation on how to change the mining settings for a classification task.

  (3) In this question, select FoodMart as your data base.

• Tasks:

   

  • Step(1) Suppose you have a classification query as follows.

          ANALYZE Gender
          ON DIMENSIONS Customers, Education Level, Product, Marital Status, Time
          FROM CUBE FoodMart-Sales
          SET Classification threshold 75.0%
          SET Noise threshold 2.0%
          SET Train Set threshold 100.0%
    

    Use the mining wizard in DBMiner to submit this classification task. Report the number of nodes, the number of leaves, and the height of the final decision tree. How many rules did you obtain finally? For the five attributes specified as above, which attribute is the most relevant one and which is the second most relevant one?

  • Step (2) For the above query, keep the same specifications you set in step (1). But in this step, dynamically change the Noise threshold until the number of rules you obtained is two. Report these two rules.
  • step (3) For the above query, set the Classification threshold to be 90.0% and the Noise threashold 0.5%. Report the number of nodes, the number of leaves, and the height of the fainl decision tree. How many rules did you obtain finally?
  • Step (4) For the above query, keep the same the specifiations. In this step, dynamically change the Noise threshold unitl there is only one rule generated from the decision tree. Report this rule.

    In general, when you change the Noise threshold, the number of nodes and the number of leaves also change. From the above steps, can you figure out what the relationship between them is in general? Explain briefly why.

• Submission:

  For each step, submit the result you obtained.

3. Comments on DBMiner (Extra 10 Marks)

Since we are still in the improving and perfecting stage of DBMiner, we welcome your any comments (Please answer this question in a separate page so that we can collect them conveniently. Thanks.).

• How do you think about the user interface of DBMiner? Show you answer specifially, such as that you think some button is not suggestive enough.
• Did you find any bugs in DBMiner? Under what conditions did you find these bugs?