DAC

Pytorch implementation of the paper Towards deep amortized clustering

Requirements

Python >= 3.5.0
torch >= 1.3.0
torchvision >= 0.5.0
tqdm
wget
sklearn

Usage

Setup

• Install required packages

pip install -r requirements.txt

• First modify ROOT variable in utils/paths.py to your working directory.

Mixture of Gaussians

• Training

python -m scripts.run \
    --model models/mog.py \
    --loss_type (min or anc) \
    --run_name (your experiment id) \
    --gpu (gpu number to work with)

• Test clustering on benchmark datasets

python -m scripst.test_cluster \
    --model models/mog.py \
    --loss_type (min or anc) \
    --run_name (your experiment id) \
    --gpu (gpu number to work with)

• Visualizing clustering

python -m scripts.vis_cluster \
    --model models/mog.py \
    --loss_type (min or anc) \
    --run_name (your experiment id) \
    --gpu (gpu number to work with) \
    --vB (number of datasets) \
    --vN (max number of points per dataset) \
    --vK (max number of clusters per dataset)

For mixture of warped Gaussians

• Same as above, but change --model argument to models/warped.py.