modelStudio provides an interactive studio for explanatory analysis of machine learning models, generating serverless HTML sites with a single line of code. It supports various black-box models and frameworks, offering a customizable dashboard with plots and descriptions for both instance-level and model-level explanations. This tool is designed for data scientists and ML engineers seeking to understand and communicate model behavior effectively.

How It Works

The package leverages the DALEX framework to create model-agnostic explainers. It computes various explanation metrics and visualizes them in an interactive dashboard. The approach integrates explanatory data analysis (EDA) with model-specific explanations, providing a comprehensive view of model performance and behavior. It supports R and Python models, with Python explainers loadable into R via reticulate and pickle files.

Quick Start & Requirements

• Install from CRAN: install.packages("modelStudio")
• Install development version: devtools::install_github("ModelOriented/modelStudio")
• Core dependencies: DALEX, DALEXtra (for R), dalex (for Python).
• Python integration requires reticulate and pickle.
• See R & Python examples for detailed usage.

Highlighted Details

• Model-agnostic, compatible with mlr3, xgboost, scikit-learn, lightgbm, keras/tensorflow, and more.
• Generates interactive, shareable HTML dashboards.
• Integrates explanatory and exploratory data analysis.
• Supports both R and Python model explainers.

Maintenance & Community

• Developed by ModelOriented.
• Citations available for JOSS and DAMI articles.
• Mentions financial support from National Science Centre (Poland) and National Centre for Research and Development.

Licensing & Compatibility

• License not explicitly stated in the README, but typically R packages on CRAN are under GPL-2 or GPL-3. Compatibility for commercial use or closed-source linking would depend on the specific license.

Limitations & Caveats

The README does not explicitly state limitations or caveats regarding supported model types beyond general compatibility, platform dependencies, or known issues. The package appears to be primarily R-focused with Python integration capabilities.