Melloddy Predictor

This open-source software is a Python package made for external data scientists without high knowledge of the MELLODDY stack to perform predictions on new data easily, from the models produced during the yearly runs. It is built on top of Melloddy-Tuner and Sparsechem to manage both data pre-processing and model inference steps. It is flexible enough to handle multiple models and data size, and predict on subset on tasks.

⚠️ The model should be compatible with sparsechem 0.9.6+. If it is not, you can convert it with the convert.py script from sparsechem.

Installation

To install the package, run in a new python 3.8+ environment:

1. Clone the repository

   git clone git@github.com:melloddy/MELLODDY-Predictor.git
   cd MELLODDY-Predictor

2. Install the package and its requirements. You can remove github if you already installed melloddy_tuner and sparsechem. You can remove doc if you don't want to build it.

   pip install -e ".[github,doc]"

3. (Optional) To be able to run the examples and the tests, download the dummy files. You can download it using make inputs. Otherwise download the archive, extract it an place the inputs folder at the root of the project.

Usage

To build the doc, run in a new terminal with your python environment:

make doc

Then in your browser, go to http://localhost:8080/melloddy_predictor/

You can see an example in example.py and run it with:

python examples/example.py

Development & Testing

Install all the requirements

pip install -r requirements-dev.txt

We use pytest for testing, you can just run the following command to run the full test suite:

make test

Note that input data will be downloaded from Zenodo when running the tests for the first time.

To set up pre-commits, you can run:

pre-commit install

to lint all files, you can run

make lint

Troubleshooting

If you want to remove the following warning:

[W ParallelNative.cpp:206] Warning: Cannot set number of intraop threads after parallel work has started or after set_num_threads call when using native parallel backend (function set_num_threads)

run:

export OMP_NUM_THREADS=1