Note: This repository is a fork of the excellent work by Subham Subba and Suman Chatterjee in their paper: "Machine learning-driven determination of key absorber layer parameters in perovskite solar cells".
We acknowledge and appreciate their contributions to perovskite solar cell research.
Our goal here is not to critique their work, but to use it as a foundation to demonstrate reproducibility best practices for educational purposes.
The original research provides insights into a predictive and interpretible model for device bandgap. Allowing for predictions on how device composition effects the bandgap, conduction band minimum (CBM) and valence band mimimum (VBM).
However, reproducing computational results can be challenging, even for robust research. This repository:
- Demonstrates how to improve code accessibility for reproducibility.
- Highlights tools and workflows to help others build on existing work.
- Acts as a learning resource for open science practices.
This repository pairs with a guided workshop to teach reproducibility best practices using the original project as a case study.
- What you’ll learn: Dependency isolation, project organization, and documentation.
- Get started: Follow the workshop guide to explore iterative improvements branch-by-branch.
To make the project more approachable for newcomers, we’ve iterated on the original code with:
| Improvement | Branch | Tools Used |
|---|---|---|
| Dependency Management | dependencies |
project.toml, uv |
| Organized Project Structure | structure |
Modular folders, path fixes |
| Documentation | docs |
README.md, docstrings |
Hyperparameter search iterations were reduced from 100 → 10 to accelerate testing. While this sacrifices result quality, it prioritizes speed, as this repository emphasizes reproducibilty over optimization.
This is not a judgment on the original work — all research code evolves! We aim to show how small changes can amplify impact.
- Original Authors: @subhamsubba for their foundational work.
- Paper: Subba and Chatterjee, “Machine Learning-Driven Determination of Key Absorber Layer Parameters in Perovskite Solar Cells.” (https://doi.org/10.1016/j.mtcomm.2024.111113).