pt-dec

This repository offers a PyTorch implementation of a variant of the Deep Embedded Clustering (DEC) algorithm. The original code can be found at vlukiyanov/pt-dec. This implementation is compatible with PyTorch 1.0.0 and supports Python 3.6 and 3.7, with optional CUDA acceleration.

This follows (or attempts to; note this implementation is unofficial) the algorithm described in "Unsupervised Deep Embedding for Clustering Analysis" of Junyuan Xie, Ross Girshick, Ali Farhadi (https://arxiv.org/abs/1511.06335).

Installing

To set up the environment for running this code, you can create a new Conda environment with python 3.11 using the following command:

conda create --name your_env_name python=3.11

Replace your_env_name with a name of your choice for the environment.

Once your env activated, run:

pip install -r requirements.txt

Troubleshooting

If Conda is not installed on your system, follow these steps to install Miniconda:

1. Download the Miniconda installer:

   cd ~
   wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

2. Make the installer executable:

   chmod +x Miniconda3-latest-Linux-x86_64.sh

3. Run the installer:

   ./Miniconda3-latest-Linux-x86_64.sh

4. Reload your shell configuration:

   source ~/.bashrc

5. Verify the installation:

   conda --version

Refer to the official Miniconda documentation for more details.

Usage

1. Data Preparation

• Transform data to Parquet format:
  Run the following script to convert your data:

  python tools/transform_to_parquet.py

• Generate embeddings:
  Create embeddings for your data with:

  python tools/generate_embeddings.py

2. Model Training

• Train autoencoder and DEC models:
  Use the main training script with customizable options:

  python tcc.py [OPTIONS]

  Key options:

  • --cuda: Use CUDA for acceleration (default: False)
  • --testing-mode: Run in testing mode (default: False)
  • --train-autoencoder: Train the autoencoder from scratch or load an existing one (default: True)
  • --sort-by-elem: Split data by "ElemDespesaTCE" and cluster each part separately (default: False)

  Example usage:

      python tcc.py --train-autoencoder False --sort-by-elem True

3. Running the Application

• Create the vector store:
  Before running the app, generate the vector store:

  python chroma_vector_store.py

• Launch the Streamlit app :streamlit: 🚀: Start the application with:

  streamlit run app.py

Other implementations of DEC

• Original Caffe: https://github.com/piiswrong/dec
• PyTorch: https://github.com/CharlesNord/DEC-pytorch and https://github.com/eelxpeng/dec-pytorch
• Keras: https://github.com/XifengGuo/DEC-keras and https://github.com/fferroni/DEC-Keras
• MXNet: https://github.com/apache/incubator-mxnet/blob/master/example/deep-embedded-clustering/dec.py
• Chainer: https://github.com/ymym3412/DeepEmbeddedClustering