Skip to content

Latest commit

 

History

History
108 lines (81 loc) · 3.25 KB

README.md

File metadata and controls

108 lines (81 loc) · 3.25 KB
Raw

README

Introduction

This repository is a simple implementation for Mean Shift(MS) algorithm. The basic idea of the algorithm I have done some records in /report/report.pdf. So, please read report.pdf first!!!

The structure of the repository

├── Algo1.py
├── Algo2.py
├── data/
├── Data_Process.py
├── Mean_Shift.py
├── README.md
├── Reference/
├── report/
├── res/
├── run.py
└── Visualize_Plot.py

Directories here include :

data/			—— The input original data(csv file)
Reference/		—— The reference paper
report/			—— My report resources
res/			—— the result of different data set/algorithms

The python file :

Algo1.py		—— The implementation of Algo1
Algo2.py		—— The implementation of Algo2
Mean_Shift.py	—— The process of the Mean Shift algorithm
Data_Process.py	—— The data processing methods
Visualize_Plot.py —— The data visualization methods
run.py			—— main function to run the algorithm

How to run

The main function is defined in the file run.py.

The main loop of the run.py is :

  1. Decide what parameters to use. The parameters include :

    1. data_path: the path of the data, e.g: data/data1.csv

    2. algo_name: to use Algo1 or Algo2, e.g: Algo1

    3. bandwidth: the bandwidth of kernel function. (If it None is set, it means it will compute bandwidth by N).

    4. threshold: the threshold of the mean shift, e.g: 0.00001

  2. By command python3 run.py

  3. The log can be shown as :

    use Algo1
get original data
get processed data
get original data
set bandwidth= 0.9336688328456105
iteration times = 1 , max_distance = 1.00001
iteration times = 2 , max_distance = 0.17686936528050956
iteration times = 3 , max_distance = 0.04682090822565989
iteration times = 4 , max_distance = 0.012449725359511938
iteration times = 5 , max_distance = 0.0033168315077783343
iteration times = 6 , max_distance = 0.0008840820836842969
iteration times = 7 , max_distance = 0.00023567080958232224
iteration times = 8 , max_distance = 6.282444587304013e-05
iteration times = 9 , max_distance = 1.674764765164474e-05
use Algo1
use Algo2
N= 100
N= 200
index = 0 centroid = [6.264581278374883, 4.835123552413281]
index = 1 centroid = [3.260652115284019, 0.9951555487518141]
index = 0 variance of cluster = 0.939911335274201
index = 1 variance of cluster = 0.4847930143675687

    The last output :

    index = 0 centroid = [6.264581278374883, 4.835123552413281]
index = 1 centroid = [3.260652115284019, 0.9951555487518141]
index = 0 variance of cluster = 0.939911335274201
index = 1 variance of cluster = 0.4847930143675687

    are the centroid and variance of the corresponding cluster.

  4. The result of points and clusters is shown in the file data_alg_bandwidth_threshold.png ， e.g: data1_Algo1_None_1e-05.png

Report

In the report directory is my Task Report, I explain my understanding of the algorithm. The report is report.pdf.

Todo list

  • Multi-threaded optimization;
  • Estimation of different parameters;