Generative Recurrent Neural Networks

This package implements the Generative Recurrent Neural Networks from the paper from Gupta et al. and combines it with the Reinforcement Learning procedures described in the paper from Olivecrona et al.

Current List of Contributors:

• Silvia Amabilino (NovaData Solutions ltd., University of Bristol)
• Michael Mazanetz (NovaData Solutions ltd.)
• David Glowacki (University of Bristol)

Installation

In order to use this package, you need the following packages installed:

• scikit-learn (0.19.1 or higher)
• tensorflow (1.9.0 or higher, but < tensorflow 2.0) or tensorflow-gpu (1.9.0 or higher, but < tensorflow 2.0)
• keras (2.2.0 or higher)
• rdkit (optional)

RDkit is only needed for the reward function in reinforcement learning.

To install, you can run the following command:

pip install git+https://github.com/SilviaAmAm/MolBot.git

Otherwise, you can clone the repository:

git clone https://github.com/SilviaAmAm/MolBot.git

and then, in your desired python environment, run:

cd MolBot
pip install MolBot/

Building the documentation

To build the documentation, these packages are required:

• nbsphinx
• sphinx_rtd_theme

In the docs folder, run:

make html

Then, open ./docs/build/html/index.html.

Usage

You can have a look at some examples in the examples folder.

Running the tests

Go to the test directory:

cd tests

Then, run all the tests:

pytest test_*