Visualization¶

The package includes optional matplotlib-based helpers for turning a ResultBundle into simple charts.

Dependency model¶

Visualization is optional. Install it through the visualization-capable extras:

uv sync --extra examples
or pip install -e ".[viz]"

Available plotting functions¶

Function	Purpose
`plot_total_metrics_bar(result_bundle)`	Plot total precision, recall, F1, specificity, micro accuracy, and row accuracy
`plot_per_feature_metrics_bar(result_bundle, metric_name="f1")`	Plot one metric per feature
`plot_entity_presence_summary(result_bundle)`	Plot entity presence summary if a compatible summary dict is present
`plot_confusion_matrix_for_classification(gold_labels, predicted_labels, class_names)`	Plot a confusion matrix from explicit labels
`plot_metric_by_group(per_feature_metrics_data_frame, group_column_name, metric_name)`	Plot mean metric by a grouping column in a pre-joined DataFrame
`save_all_charts_to_report(result_bundle, output_directory_path)`	Save standard charts into a timestamped report folder

General behavior¶

The plotting helpers are intentionally simple:

one chart per figure
deterministic sorting for per-feature and grouped plots
y-axis constrained to [0.0, 1.0] for score-based charts
empty or invalid inputs usually produce a figure with a clear note instead of failing

Report generation¶

save_all_charts_to_report(...):

ensures the output directory exists
creates a timestamped folder named report_<timestamp>
saves standard charts as PNG files
returns a mapping from chart name to file path

Current standard outputs:

total_metrics.png
per_feature_f1.png
entity_presence.png when available

Current implementation notes¶

Entity presence summary keys¶

plot_entity_presence_summary() currently looks for summary keys named:

precision
recall
f1

The multi-entity evaluator currently stores:

precision_entities
recall_entities
f1_entities

So if you pass the raw ResultBundle from MULTI_ENTITY evaluation directly, this chart may not display the intended values unless you adapt the summary keys first.

Matched-pairs helper expectations¶

extract_labels_for_feature() expects a matched_pairs_data_frame containing columns named like:

<feature_name>_gold
<feature_name>_pred

The current evaluator's matched_pairs_data_frame contains only indices and similarity scores, so that helper is mainly useful with custom or enriched result bundles.

Example¶

from extraction_testing.visualization import (
    plot_total_metrics_bar,
    plot_per_feature_metrics_bar,
    save_all_charts_to_report,
)

fig_total = plot_total_metrics_bar(result_bundle)
fig_f1 = plot_per_feature_metrics_bar(result_bundle, metric_name="f1")
paths = save_all_charts_to_report(result_bundle, "./_viz_out")