PyCaret 4.0 plot library — Plotly-native, framework-agnostic.
Replaces the legacy plot_model API (deleted in s46). Each plot
function takes a fitted estimator (typically a sklearn / sktime
Pipeline) plus data and returns a plotly.graph_objects.Figure.
The figure is the universal currency: notebooks call .show(), the
FastAPI layer calls .to_dict() for HTTP transport, the React UI
consumes the JSON via react-plotly.js.
Module layout:
classification— confusion matrix, ROC, PR, calibration, threshold, lift, gain.regression— residuals, prediction-error, learning curve.feature— feature importance + SHAP (s48).clustering— silhouette, elbow, embedding (s49).anomaly— score histogram, anomaly map (s50).time_series— decomposition, forecast fan, residual diagnostics (s51).eda— distributions, correlation, missingness (s52).
Design tokens align with the React UI's ink-* / accent-* /
success-* / danger-* palette so plots look at home in the
dashboard without per-screen restyling.