FAQ

What model types does ModelDoctor support?

ModelDoctor officially supports scikit-learn compatible estimators (BaseEstimator). This includes models from Scikit-Learn, XGBoost, LightGBM, and CatBoost (when using their sklearn wrappers).

Does it support regression?

Yes. ModelDoctor automatically infers whether the task is classification or regression and routes the context to the appropriate diagnostic logic within the Doctors.

Why are SHAP charts missing from my dashboard?

The ExplainabilityEngine requires the shap library. If it is not installed, ModelDoctor gracefully falls back to scikit-learn's native permutation importance and skips SHAP rendering. Install it via pip install "modeldoctor[shap]".

The dashboard won't open when I call report.dashboard()

Some operating systems or notebook environments restrict the opening of temporary HTML files. You can bypass this by manually saving the file: report.save_html("report.html") and opening it in your browser.

Is ModelDoctor slow on large datasets?

ModelDoctor lazily computes expensive metrics (like cross-validation or permutation importance). However, if a Doctor requests these metrics on a dataset with millions of rows, execution can take several minutes. We recommend passing a stratified sample (e.g., 50k rows) to md.diagnose() for rapid iteration.

Can I compare two models side-by-side?

ModelDoctor evaluates a single model pipeline per diagnose() call. To compare models, generate two reports and view the HTML dashboards in separate browser tabs.

Are you planning to support PyTorch/TensorFlow?

Yes, deep learning support is on the roadmap. Neural networks present unique diagnostic challenges (vanishing gradients, catastrophic forgetting) which we are currently building dedicated Doctors for.