Overview

This repository contains a simple implementations of z-score data standardization and k-means clustring algorithms created using CUDA framework and C++ language. Created program runs both parallel and single-threaded versions of aforementioned algorithms to compare the performance.

Running the project

To build the project one should entry the project directory and execute following commands:

mkdir build
cd build
cmake ..
make

In addition, one can decide to build the project using either Release of Debug configuration. To achieve this, one should add -DCMAKE_BUILD_TYPE=Release or -DCMAKE_BUILD_TYPE=Debug option. To run this project, one should type:

./CUDA_Data_Mining_cpp <number_of_clusters> <number_of_iterations> <data_file> <columns_to_omit>

There are four command line arguments this program takes, but the last one could be optional, depending on used dataset. They have a following meaning:

• <number_of_clusters> - number of clusters and initial centroids program should create
• <number_of_iterations> - number of iteration of the K-Means algorithm that should be performed
• <data_file> - path to file with data
• <colums_to_omit> - indexes of dataset columns that should not be loaded from file (useful when the dataset contains non-numerical data, that can not be handled by this implementation)

For example, to create 3 clusters in 100 iteration using the iris dataset, one should type:

./CUDA_Data_Mining_cpp 3 100 ../data/iris.data 4

Repository structure

This repository contains following directories and files:

• common - contains C++ code responsible for common functionalities, for example: parsing command line arguments, loading dataset from file, handling CUDA errors and choosing initial centroids
• cpu - contains C++ code with single-threaded implementation of data standardization and k-means algorithms
• gpu - contains C++ code with CUDA implementation of data standardization and k-means algorithms
• python_scripts - contains utility scripts for generating test data and plotting results
• data - contains example dataset (Iris) used for testing
• main.cu file - application entry point
• CMakeLists.txt - CMake file that I used to build and test the project