This repository contains source code for the paper Combining Latent Space and Structured Kernels for Bayesian Optimization over Combinatorial Spaces accepted at NeurIPS-2021.

• Running the code

  • Setting up conda environment from https://github.com/cambridge-mlg/weighted-retraining because we use their pre-existing VAE models. See 'Install dependencies' section.

  • Download datasets by following the 'Set up Data' section in the above repository.

  • Relative imports are setup as if LADDER directory's code is in weighted-retraining/weighted_retraining/opt-scripts/.

  • To run LADDER for chemical design task: python ladder_chemical_design.py

  • To run LADDER for arithmetic expression optimization, some things need to be set up in above weighted-retraining environment

    • Add a comma in line 34 of weighted-retraining/weighted_retraining/train_scripts/train_expr.py
    • unzip eq2_grammar_dataset.zip in weighted-retraining/assets/data/expr
    • run python ladder_expression.py

This code builds upon the existing code provided by authors of Sample-Efficient Optimization in the Latent Space of Deep Generative Models via Weighted Retraining. We thank the authors for their code.

Please consider citing both the papers if you use this work:

• Austin Tripp, Erik Daxberger, and Jos{'e} Miguel Hern{'a}ndez-Lobato. Sample-Efficient Optimization in the Latent Space of Deep Generative Models via Weighted Retraining, NeurIPS 2020.
• Aryan Deshwal and Janardhan Rao Doppa. Combining Latent Space and Structured Kernels for Bayesian Optimization over Combinatorial Spaces, NeurIPS 2021.