MLRAP (Machine Learning Regression Analysis Package) is designed to assist materials scientists in constructing regression models and establishing structure-property relationships. This software offers a user-friendly command-line interface and automates the machine learning workflow in Python. It encompasses modules for data preprocessing, feature engineering, model evaluation, model optimization, and model interpretation. This automation significantly reduces the workload for users and enhances the efficiency of materials research and development.
MLRAP supports Python 3.8+. It is recommended to install it within a virtual environment using virtual environment management tools.
First create a conda environment: Install miniconda environment from https://conda.io/miniconda.html based on your system requirements. Then, create a virtual environment:
conda create -n mlrap python==3.8.0
conda activate mlrap
Extract the compressed archive to a directory, and then execute the following command:
python setup.py install
We highly recommend installing in editable mode:
pip install --editable .
After installation, execute the command mlrap -h to view the help interface.
Usage: mlrap [OPTIONS] COMMAND [ARGS]...
Machine learning regression analyse packages
Options:
-h, --help Show this message and exit.
Commands:
run Global run base config file.
subrun Run step by step.
If you have already prepared your dataset (a CSV file containing chemical formulas and target properties), simply execute the command mlrap run train in the directory where your dataset is located. MLRAP will automatically perform feature engineering, hyperparameter optimization, and model training for you. Process data will be output in CSV format, and five images will be generated as follows:
During hyperparameter optimization, we employ Bayesian hyperparameter optimization, where the best hyperparameters are selected for the final model training.