yczheng0 / kmm-python

COMP 5331 group project

KMM-Python

COMP 5331 group project

Reference: Paper , Matlab code

Prerequisite

• OS: Linux & Mac OS(tested)

• Python 3.7.4

• pipenv

Setup

cd <work-dir>
git clone https://github.com/SimonZYC/KMM-Python.git
cd KMM-Python

pipenv install

Run

Toy Sample (For demonstration)

cd <Path-to-KMM>
./bin/test.sh toy

Test All (very slow)

cd <Path-to-KMM>
./bin/test.sh all

Custom

cd <Path-to-KMM>

pipenv run python bin/test_kmm.py -d data/overlap/s4.data -g data/overlap/s4.ground

Source Code Description

KMM-Python

• bin:
  • test.sh, test_KMM.py: test scripts
  • batchPlot.py: create a figure of the results
  • batchRun.py: execute serveral clustering algorithms together

​

• KMM: KMM code directory:

  • KMM.py: main function, control the iteration process.
  • csbg.py: Solve the optimization problemm
  • ConstructA_NP.py: create sparse matrix of distance between points
  • gen_nn_distanceA.py: calculate distance between points
  • sqdist.py: computes a rectangular matrix of pairwise distance between points in A (given in columns) and points in B
  • SVD2UV.py: transformation

• other: Implementaion of k-means, kernel-k-means, spectral:

  • k_means.py: the code of k-means algorithm
  • kernel_k_means.py: the code of kernel k-means algorithm
  • spectral.py: the code of spectral clustering algorithm

• utils: some tool functions

  • affinity.py: the affinity function and pruning function used in the spectral clustering algorithm
  • analysis.py:
  • kernels.py: the kernel functions used in the kernel k-means algorithm
  • LoadData.py: the functions used to load data points and related labels
  • utils.py: the functions used to calculate NMI, purity and distance

​