AutoGMM

This repo reproduces the results in "AutoGMM: Automatic Gaussian Mixture Modeling in Python" (https://arxiv.org/abs/1909.02688)

Installation

The algorithm is located in the branch of https://github.com/neurodata/graspy. So, first install graspy through github, then navigate to the autogmm branch usin the following commands:

git clone https://github.com/neurodata/graspy
cd graspy
git checkout autogmm
python3 setup.py install

or

git clone https://github.com/neurodata/graspy
cd graspy
git checkout autogmm
pip install -e .

The algorithm is located in graspy/graspy/cluster/autogmm.py. After autogmm is installed, you can run the scripts below.

To run the R scripts, you will need to install R and the mclust library (we use version 5.4.2 in the paper). We recommend the RStudio IDE https://www.rstudio.com/. Users may need to "Set Working Directory" to "Source File Location," for the scripts to find find relative paths correctly.

Directories

complete_experiments

These files reproduce Table 2, Figures 1-3, and Figure 5. They run the clustering algorithms on the complete datasets. Instructions within.

subset_experiments

These files reproduce Figure 4. They run the clustering algorithms on the subsets of the data. Instructions within.

option_runtimes

These files reproduce Figure 6. Instructions within.

brute_cluser_graspyclust.py - implementation of graspyclust
make_gmix.py - script that was used to make data/synthetic.csv
./data/ - contains the datasets that was used in the paper