BiRLNN: Bidirectional Reinforcement-Learning Neural Network for Constrained Molecular Design

This zip file contains the training dataset and Python code necessary to reproduce the results in the manuscript BiRLNN: Bidirectional Reinforcement-Learning Neural Network for Constrained Molecular Design. 

The three zip files contain:

• code.zip: the source code for the BiRLNN package
• data.zip: the complete training datasets used to train the models
• evaluation.zip: the generated results presented in the paper

Usage:

1. unzip all three files
2. If only need to check the existing results, they can be found under the evaluation/ folder
3. To reproduce the results, place data/ and evaluation/ under code/. Set up conda environment using the birlnn.yml file. The python and bash scripts (placed in code/scripts and code/model) should be runable from within the folder