visualization/

This package turns the numbers produced by metrics and statistics into publication-quality figures. Reach for it at the end of an experiment, when you already have predictions or a score matrix and want to see — or publish — what they say.

Two families of plots live here:

Single-model diagnostics — you have one fitted model and its predictions on a test set:

• plot_confusion_matrix — which classes

get mistaken for which.

• plot_roc_curve /

plot_pr_curve — threshold-free ranking quality; the PR curve is the one to trust on imbalanced data.

• plot_learning_curve — whether more

training data would still help.

• plot_tree — the structure of a fitted

decision tree.

Multi-model comparison — you have a (n_datasets, n_algorithms) score matrix from a benchmark:

• plot_critical_difference — the

Demšar (2006) rank diagram: which algorithms are statistically distinguishable, not merely different on average.

• plot_ranking_table — the raw numbers

plus per-dataset ranks.

• plot_boxplot_comparison — score spread

per algorithm.

• plot_heatmap — the whole score matrix

at a glance.

All plotting functions share the same conventions: matplotlib is imported lazily, so they raise ImportError if it is unavailable; each one calls matplotlib.pyplot.show() before returning, and writes a 300-dpi PNG when a save_path is supplied. Styling comes from a shared internal stylesheet whose palettes (PALETTES) and semantic colour tokens (SEMANTIC_COLORS) are re-exported here, together with the helpers apply_style, reset_style, get_colors, setup_figure and style_axis for building your own matching figures.

>>> import numpy as np
>>> from tuiml.evaluation.visualization import plot_confusion_matrix
>>> y_true = np.array([0, 1, 1, 0, 2, 2, 1])
>>> y_pred = np.array([0, 1, 0, 0, 2, 1, 1])
>>> cm = plot_confusion_matrix(y_true, y_pred)   # doctest: +SKIP