New features
Bayesian-FPDE. Build a posterior overlambda_hyb candidates instead of picking a single value. Call FPDEEngine.select_bayesian_lambda to get a posterior mean, MAP estimate, and credible interval. Then use explain_one_bayesian or explain_batch_bayesian to produce attributions under the posterior mean. Reach for this when you want model-averaged Hyb-FPDE explanations with uncertainty over the mixture weight. See Select lambda and the API reference.
Plotting helpers. A new optional plotting module visualizes FPDE attributions with bar, waterfall, beeswarm, and scatter plots. It also covers prototype similarity distributions and perturbation curves. Install it with pip install "fpde[plot]", then import from fpde or fpde.plots. See the quickstart for examples.
Updates
Example notebook refresh. A focusedplot_fpde_example.ipynb replaces the legacy MNIST notebook. It walks through fitting an engine, generating explanations, and producing the new plots end to end.