SFE-Finder is a quantum machine learning project that aims to predict Stacking Fault Energy (SFE) values for various elements using their physical properties. This project leverages quantum computing techniques, specifically using PennyLane, to create a hybrid quantum-classical model for SFE prediction.
- Utilizes quantum circuits for feature processing
- Implements a hybrid quantum-classical model
- Predicts SFE values based on electronegativity, bulk modulus, and atomic volume
- Provides data preprocessing and model training scripts
- Python 3.7+
- PennyLane
- NumPy
- Pandas
- Scikit-learn
The project uses a dataset (qml_training-validation-data.csv) containing the following features for various elements:
- Electronegativity (el_neg)
- Bulk modulus (B/GPa)
- Atomic volume (Volume/A^3)
- Stacking Fault Energy (SFE/mJm^-3) as the target variable
- Model Architecture
The quantum part of the model consists of:
- Input encoding using RY rotations
- Parameterized quantum circuits with RX, RY, and RZ rotations
- Measurement of expectation values of Pauli Z operators
The classical part includes:
- Data preprocessing and scaling
- Optimization using gradient descent
- Cost function based on Mean Absolute Error
The model's performance is evaluated using Mean Absolute Error (MAE) and R-squared score. Current results and any visualizations can be found in the notebook. The accuracy is 96%.
Contributions to improve the model's accuracy, efficiency, or to extend its capabilities are welcome. Please feel free to submit pull requests or open issues for discussion.