mentor.reporting

Utilities for inspecting and visualising mentor checkpoint files without instantiating the model.

mentor.reporting.get_report_str(path, terminal_colors=True, verbose=False, render_colors=None)

Generate a human-readable text report for a mentor checkpoint file.

Loads the checkpoint with map_location="cpu" so no GPU is required. Does not instantiate the model class — all information is derived directly from the serialised data.

Parameters:

• path (str) – Path to a .pt checkpoint file created by save().

• render_colors (bool, optional) – If True, the returned string contains ANSI colour escape codes suitable for terminal display. Defaults to False.

• terminal_colors (bool)

• verbose (bool)

Returns:

Multi-line report covering: file size, model class, architecture statistics, training and validation history, software provenance, plottable metric names, inference state inventory, output schema, preprocessing info, and checkpoint contents.

Return type:

str

Examples

>>> from mentor.reporting import get_report_str
>>> print(get_report_str("model.pt"))
>>> print(get_report_str("model.pt", render_colors=True))

mentor.reporting.plot_history(values, paths, overlay=False)

Plot training/validation history from one or more checkpoint files.

Checkpoints are loaded with map_location="cpu"; no GPU is required. Each file gets a distinct colour; each metric a distinct line style. Vertical dashed lines mark the best-epoch for each file when available.

Parameters:

• values (list[str]) – Metric names in split/metric form, e.g. ``[“train/loss”, “validate/acc”]”. Pass an empty list to auto-discover all available metrics (union across all files).

• paths (list[str]) – One or more paths to .pt checkpoint files.

• overlay (bool, optional) – If True, all metrics and files share a single axis. If False (default), one subplot per metric with all files overlaid on each subplot.

Returns:

The composed figure. Call fig.savefig(...) or plt.show() to display it.

Return type:

matplotlib.figure.Figure

Examples

>>> from mentor.reporting import plot_history
>>> fig = plot_history([], ["run1.pt", "run2.pt"])
>>> fig.savefig("comparison.png", dpi=150, bbox_inches="tight")