zhifengkong / 203Project

On the EM Algorithm for Gaussian Mixture Models with Adversarial Regularization

203Project

On the EM Algorithm for Gaussian Mixture Models with Adversarial Regularization

Datasets index

• Multi-adv-0: Missing outside, 4 adv
• Multi-adv-1: Missing 4 strips ("#" shape), 4 adv
• Multi-adv-2: Missing 4 strips ("#" shape), 8 adv
• Multi-adv-3: Missing 2 strips ("+" shape), 8 adv

Dependencies

Python 3, numpy, scipy, (pickle for saving results, tqdm for logging, and matplotlib for vis)

Example Runs

standard

python em_train.py --algo penalized --output_dir results/multi-adv-test --dataset_file multi-adv-0/data_multi_adv.npz

online

./online_sample.sh

Visualization

Dataset

python visualize.py multi-adv-0/data_multi_adv.npz  --save_path multi-adv-test/dataset.jpg

Results

python visualize.py multi-adv-test/results.p  --save_path multi-adv-test/em.jpg

Notice

• Don't modify the Results.ipynb directly; copy it to a new file named Results-xxx.ipynb if changing is needed (it will be in .gitignore).
• Please put datasets, figures and results into corresponding folders (refer to the datasets index).