structboost.plot_training_diagnostics

structboost.plot_training_diagnostics(report, *, figsize=(13.0, 9.0), log_scale=True)[source]

Plot a TrainingReport as a six-panel diagnostic figure.

Parameters:

  • report – A TrainingReport, a fitted BAE (its training_report is used), or an AnnData carrying uns["bae"]["training_report"].

  • figsize (tuple[float, float]) – Figure size in inches.

  • log_scale (bool) – Use a log y-axis for the convergence and gradient panels, whose values span orders of magnitude.

Returns:

matplotlib.figure.Figure – The figure. Panels are: reconstruction loss, per-iteration contribution of the encoder and decoder halves, encoder convergence, selected genes, gradient norms, and boosting fit quality.

Raises:

ValueError – If no training report is available (diagnostics=True was not set).