Silhouette clustering coefficient with PyOpenCL

(PA I project PyOpenCL)

Project to Paralel Algorithms I course. Aim is to create parallel version of count of Silhouette clustering coefficient for some clustered Data

Requirements

• OpenCL driver based on your device architecture (AMD, Nvidia, Intel etc.)
• Python 2.6+
• PyOpenCL
• Numpy
• MingWPy
• Mako

File stucture

Project contains 3 exectutable standalone main files.

• main.py - contains non-parallel version of algorithm
• main_parallel.py - first version of paralel computation (prove of concept)
• main_parallel2.py - Second version with slight improvement

Programs - File contains kernel programs used for parallel computation

• distance_matrix.cl
• silhouette_1.cl - first version
• silhouette_2.cl - second version

Data - File contains sample data

Run

Each file have at the top FILE_PATH constant which contains path to sample data Sample data have to be in JSON file in following structure:

{
  "clusters":[
    [[1,2], [4,5]],
    [[5,4], [55,43]],
  ]
}

Key clusters contains array of clusters. Each cluter contains array of individual members of cluster. Members are represented as arrays.

python main_parallel.py