The main report of this work is the pdf project_report.pdf.

Simply build a virtual environment using virtualenv or conda and install the requirements.

virtualenv env-nmf-clustering
source env-nmf-clustering/bin/activate
pip install -r requirements.txt

To avoid any importation error we have a flat organization (all the files are in the same main folder).

Here are the description of each file.

dataset.py: functions to build the parametric dataset.

kmeans.py: KMeans class and related functions like KMeans++.

nmf.py: NMF classes.

visualization.py: Utility function to plot our figures using plotly.

All the outputs presented in the report are from Python scripts found here. They can all be launched using Python CLI. Lauchning them with --help will print a description of the different option. But they can be launched as it, with coherent default.

If you want the figures to be printed on screen, use the --show arguent (not default).

dataset_experiment.py: A small script to play with the parametric dataset.

kmeans_pp_experiment.py: The experiments related to KMeans++ initialization.

nmf_experiment.py: The experiments related to 2-NMF and 3-NMF.

data_embedding_experiment.py: A small experiments in which we embed the 2D data into a high dimensional space.