Introducing ChemBoost

ChemBoost: SMACT-Based Intermetallics Classification Package

Overview

This release introduces ChemBoost, a machine learning package designed to enhance SMACT's filtering capabilities by classifying intermetallic compounds as metals or non-metals. The package leverages XGBoost and materials science features to provide accurate predictions based on compositional properties. It is still a work in progress, and will evolve as time goes by, feel free to reach out with ideas and potential solutions to any challenges encountered.

Key Features

• Advanced Feature Engineering:

  • Valence Electron Count (VEC) calculations
  • Electronegativity difference metrics
  • Atomic concentration analysis
  • Magpie feature integration

• Robust Model Training:

  • XGBoost classifier with GridSearchCV optimization
  • Hyperparameter tuning
  • Cross-validation support

• Comprehensive Visualization:

  • Confusion matrix plotting
  • SHAP value analysis for model interpretation
  • Feature importance visualization
  • Distribution analysis of key features

Package Structure

chemboost-classification/
├── src/ # Core package code
│ ├── data/ # Data loading utilities
│ ├── features/ # Feature engineering
│ ├── models/ # ML model implementation
│ └── visualisation/ # Plotting utilities
├── notebooks/ # Analysis notebooks
└── data/ # Dataset storage

Dependencies

• Python 3.11+
• XGBoost 2.1.3
• scikit-learn 1.5.2
• matminer
• pymatgen
• SMACT 2.8+

Usage

The package can be installed and run using:

pip install -e .
python run.py

Technical Details

• Implements a modular, maintainable codebase
• Includes comprehensive documentation
• Provides both command-line and API interfaces
• Features extensive error handling and input validation

Testing

• Tested with Matbench experimental bandgap dataset
• Validated against known metallic/non-metallic compounds
• Cross-validated model performance metrics included

Future Enhancements

• Developing more robust feature handling to help improve the filter's capabilities and further separate metals and non-metals accurately
• Integration with SMACT's main filtering pipeline
• Support for more complex chemical environments
• Extended visualization options

Notes

• Requires scikit-learn <1.6 due to API changes that are currently incompatible with xgboost
• Includes sample data for immediate testing
• Documentation available in README and docstrings