chemprop-contrib

Community-contributed extensions to the Chemprop machine learning package

Usage

Installation

chemprop-contrib can either be pip installed directly

pip install chemprop-contrib

or installed as an optional dependency when installing Chemprop

pip install chemprop[contrib]

Then, the contributed packages can be imported in Python as

from chemprop_contrib import <package_name>

Each package includes tests, which can be used to demonstrate the use of the package.

Available Packages

• moe_regressor: Implements the "Adaptive Mixture of Local Experts" model for regression.
• mcp: Model context protocol for connecting Chemprop with Large Language Models (AI).
• set2set: Chemprop-compatible implementation of the Set2Set aggregation function.
• sklearn_integration: Scikit-learn style wrappers around Chemprop training and predicting routines.
• mixtures: Order invariant architectures that accept any number of inputs. Useful for mixtures of molecules.

Contributing

All community-contributed packages should fork the repository and in a branch on the fork, make a new directory under the chemprop_contrib directory with their name. Then add the following:

• License for the code (preferably MIT)
• All associated code files
• Data files, if needed, and their provenance
• Documentation (perhaps a Notebook (*.ipynb) or README.md) with instructions on how to use code contributed
• Tests demonstrating that the code works as intended

If your package has additional dependencies, add them to the list of optional dependencies in the pyproject.toml with the same name as your directory.

Once the package is ready, make a pull request to the main chemprop-contrib repository for a brief review. We provide a template for these pull requests.

Examples

For packages which do not require any new dependencies, see moe_regressor as an example.

For packages which do require additional dependencies, see mcp as an example. Note that one must edit the pyproject.toml and add these dependencies, as mentioned in Contributing.