oximachinerunner

Installation

On macOS you need to run brew install libomp first to enable multithreading for the XGBoost library.

Ideally, you install everything in a clean environment, e.g., using conda

conda create -n test_oximachine_runner python=3.7 -y

Then activate with conda activate test_oximachine_runner.

Latest stable release

pip install oximachinerunner

Note that the installation requires significant (>500 MB) storage space since the ensembles use k-nearest neighbors models.

Development version

pip install git+https://github.com/kjappelbaum/oximachinerunner.git

Usage

Loading the model

from oximachinerunner import OximachineRunner
runner = OximachineRunner()

The OximachineRunner can be initialized with a modelname from runner.available_models.

By default, models will be automatically downloaded (upon first use) if there are not yet in the correct folder:

/Users/kevinmaikjablonka/opt/miniconda3/envs/test_oximachine_runner/lib/python3.7/site-packages/oximachinerunner/assets/all_202000830/classifier.joblib are not exist or md5 is wrong.
Download file from https://www.dropbox.com/s/lc2z4abaycjbbe1/classifier.joblib?dl=1
2.9% of 527.44M

To disable this behavior of, set OximachineRunner(automatic_download=False) and manually download your model, e.g. using a function from the utils module.

Predicting oxidation states

The run_oximachine function accepts pymatgen.Structure, ase.Atoms and str as well as os.PathLike. The latter two are expected to be paths to a file that is then parsed with pymatgen.

runner.run_oximachine('oximachinerunner/assets/ACODAA.cif')

The function prints for how many sites it will run the model.

featurize.py: iterating over 6 metal sites

It returns an OrderedDict with the fields:

• metal_indices: A list of indices of the metal sites
• metal_symbols: A list of symbols of the metal atoms
• prediction: A list of oxidation state predictions
• max_probabs: For each metal site the maximum confidence of all 4 models.
• base_predictions: For each metal site a list of the oxidation state predictions for each of the 4 models.

Development setup

git clone https://github.com/kjappelbaum/oximachinerunner
pip install -e .[dev]

Reference

Jablonka, Kevin Maik; Ongari, Daniele; Moosavi, Seyed Mohamad; Smit, Berend (2020): Using Collective Knowledge to Assign Oxidation States. ChemRxiv. Preprint. https://doi.org/10.26434/chemrxiv.11604129.v1