victorjcarolino / k-means_clustering

K-Means

Execution Data

https://docs.google.com/spreadsheets/d/1ZcLFo0Y49qIlUCRqSpkb2Dtbx0JHtt5JW1wI-EmIYKQ/edit?usp=sharing

Description

Kmeans-Serial

• Base code provided and referenced from https://github.com/marcoscastro/kmeans
• Keeps track of:
  • points + what cluster they are in
  • clusters + what points are in them
• Does the summation and mean of the cluster means at the end of each point association

Better-Kmeans-Serial

• Updated code from the aforementioned reference
• Keeps track of:
  • points + what cluster they are in
  • clusters + the intermediate sums from point addition and removal
• Does the summation of points to the mean as points are iterated
  • Mean is computed sequentially at the end

Kmeans-Parallel

• Parallelized code from the Better-Kmeans-Serial
• Parallelization is done over the points rather than over the clusters due to a greater chance of sufficient parallel slack
• Data contention is handled through the use of thread local storage
• Keeps track of:
  • points + what cluster they are in
  • clusters + the intermediate sums from point addition and removal
  • thread local storage handles the summation of point additions and the number of points added to a cluster as well as the cluster switches made by points
• The thread local storage resolves the intermediate sums to the appropriate cluster and resolves the addition or removal of points to ensure proper calculation of the mean

Datasets

• Apple Quality
  • K: 8
  • Total Points: 3999
  • Total Features: 7
• Beans
  • K: 17
  • Total Points: 13611
  • Total Features: 16
• Big One
  • K: 32
  • Total Points: 100000
  • Total Features: 3
• Birch
  • K: 100
  • Total Points: 87996
  • Total Features: 2
• Gaussian Distribution
  • K: 9
  • Total Points: 10125
  • Total Features: 15
• HW Set 1
  • K: 2
  • Total Points: 7
  • Total Features: 2
• HW Set 2
  • K: 3
  • Total Points: 150
  • Total Features: 4
• Pulsar
  • K: 2
  • Total Points: 17898
  • Total Features: 8

Running

• To execute the program:
  • run: sh run.sh datasets/[dataset-name].txt [output-subdirectory]
  • the outputs of the execution will be found in:
    • serial: outputs/[output-subdirectory]/kmeans-serial.txt
    • better serial: outputs/[output-subdirectory]/better-kmeans-serial.txt
    • parallel: outputs/[output-subdirectory]/kmeans-parallel.txt