Scalable Sampling for Nonsymmetric Determinantl Point Processes

Python Implmentation for Scalable Sampling for Nonsymmetric Determinantl Point Processes

• The code files are organized for (1) sampling nonymmetric DPPs (NDPPs) and (2) learning with orthogonality constraints.
• The code is based on https://github.com/cgartrel/nonsymmetric-DPP-learning/ (Scalable Learning and MAP Inference for Nonsymmetric Determinantal Point Processes, ICLR 2021)

Usage

Experiments for learning orthogonal NDPPs:

• First, download the datasets

  bash download.sh
  

• To run the ONDPP learn with UK Retail dataset,

  cd ./learning
  bash script_ondpp.sh
  

Experiments for scalable sampling from NDPPs:

• To run synthetic dataset,

  cd ./sampling
  python run_synthetic.py
  

• This will run the Cholesky-based sampling and tree-based rejection sampling

• Parameters in ``run_synthetic.py'' (e.g., ground set size) can be changed

• Models learned from real-world datasets also can be used for sampling (parameters are saved in ../learning/saved_models/)