Description pending...

• [X. Peng et al., IJCAI 2016, IEEE 2018] - Deep Subspace Clustering with Sparsity Prior
  • Author code
  • Naive implementation
• [P. Ji et al., NIPS 2017] - Deep Subspace Clustering Networks
  • Author code

Download or git clone the repository.

• Tensorflow - installed in a working Python environment such that import tensorflow throws no errors
• Matlab - has an activated license, can be launched from desktop
• OR Octave - can be launched via octave in console

Requires a working and activated version of MATLAB.

Instructions taken from here.

Launch MATLAB and run:

>>> matlabroot

Run the following on the command line, substituting the previous result into the first line. Make sure that python references a python version that has tensorflow installed:

cd matlabroot
cd extern/engines/python
python setup.py install

If the last line fails due to denied permissions, point it to a directory that you have write permissions for. You can delete this directory afterwards:

python setup.py build --build-base=/full/path/to/temp install

Check the installation by running in Python:

>>> import matlab.engine
>>> eng = matlab.engine.start_matlab()
>>> eng.isprime(37)

A viable substitute for not having MATLAB. Requires a working version of GNU Octave.

Install oct2py. Make sure that pip is tied to a python version that has tensorflow installed:

pip install oct2py

Check the installation by running in Python:

>>> from oct2py import octave
>>> octave.isprime(37)

Indicate that you'll be using Octave. If on linux, add this to your .bashrc. If on windows, use SETX:

export ENGINE_CHOICE=OCTAVE

If running kmeans in the Octave console says that it's undefined, you need to install the statistics package:

sudo apt-get install octave-statistics

Then, add the line pkg load statistics to SpectralClustering.m

If you don't have access to sudo, do the following instead. Make sure to substitute in the correct version number:

apt-get download octave-statistics
dpkg -x octave-statistics_1.2.4-1_all.deb dir

octave:1> addpath(genpath("~/path/to/dir/usr/share/octave/packages/statistics-1.2.4"))
octave:2> savepath

Verify that kmeans works:

octave:3> kmeans(randn(10, 2), 2)

Through pip or conda (some require -c conda-forge), install the following:

• scikit-image
• scikit-learn
• scikit-optimize
• munkres
• scipy
• matplotlib
• mnist
• pathlib
• pillow
• pympler

Download code for SSC using ADMM. Unzip the contents into SSC_ADMM_v1.1, located at the root of the repository.

Download and unzip any desired datasets. By default, the loading methods look for a data folder at the root of the repository.

• Extended YaleB (cropped images)
• Coil20 (processed)
• MNIST will be automatically downloaded the first time load_MNIST is run
• CIFAR10 (python version)

Alternatively, you can just load the already preprocessed matrices from saved.