Introduction to Machine Learning Program Assignment #2 Solution

Description

This program assignment aims to help you understand the K -means and Kd-tree implementation.

9. K-means Problem

You will get a dataset (data_noah.csv). It is Noah Syndergaard’s pitches that have been tracked by the PITCHf/x system in the MLB Regular Season.

You have to do the following:

1. Dataset including 1321 number of instances with many attributes.

2. Don’t use the library related to K-means. (i.e. Construct a K-means function by yourself).

3. Use Attribute x (horizontal movement) and y (vertical movement) to partition 1322 pitches into 3 clusters.

4. 3 clusters will represent FF (four-seam fastball), CH (changeup) and CU (curveball).

5. Construct a cost function to check the accuracy of pitch types.

6. Generate a figure to show the result of K-Means clustering. For example:

7. Try to use another two or more attributes (like speed) to partition.

Don’t worry whether the accuracy is high or not!

7. Try to explain why k = 3 is the best, and write in your report.

7. Show your code, accuracy, the reason of k = 3 and the result of K-Means clustering (figure) in your report.

34. • If you are interested, you can get more information of pitches from brooksbaseball. (http://www.brooksbaseball.net/landing.php?player=592789)

34. Kd-tree Problem

You will get a set of points (points.txt) in the unit square (all points have x-coordinates and y-coordinates). You have to build a 2d-tree.

You have to do the following:

1. You can use the library related to Kd-tree.

2. Draw a 2d-tree divides the unit square (Use two colors). For example:

3. Show your code and the result of 2d-tree (figure) in your report.

60. • If you are interested, you can construct a Kd-tree function by yourself.

60. • • Calculate the variance of this two dimensions and select the big one as axis-aligned splitting planes.

60. • • Then, sort points in the given set and choose median as pivot element where you should split.

60. • • As one moves down the tree, one cycles through the axes used to select the splitting planes. (For example, in a 2-dimensional tree, the root would have an x-aligned plane, the root’s children would have y-aligned planes, the root’s grandchildren would have x-aligned planes, and so on.)

60. Report & Scoring

This is a team-based program assignment, so one team should only submit one report and one source code to E3.

The report should contain the following:

1. What environments the members are using (5%)

2. K-means code (30%)

3. Cost function and accuracy (15%)

4. The result of K-Means clustering (10%)

5. Use another two or more attributes to partition and the reason of k = 3 (10%)

6. Kd-tree code (15%)

7. The result of Kd-tree (15%)

1. C / C++ / Java / Python / Matlab are allowed to use. For visualization, Excel or other programs are allowed.

2. Report format should be PDF.

3. Attach your code when you are submitting.

4. No cheating and plagiarizing.

5. Delay：Your score *= 0.8