skfair.comparison

from skfair.comparison import ComparisonReport

skfair.comparison.ComparisonReport

Visualization report for fairness method comparison results.

Parameters:
  • results_df (DataFrame) –

    DataFrame with columns dataset, method, classifier, plus one column per metric (e.g. accuracy, spd). Optional {metric}_std columns are preserved but not plotted.

  • metrics (list of str, default: None ) –

    Explicit list of metric column names to use. When None, metrics are auto-detected from the DataFrame columns.

Examples:

>>> report = ComparisonReport(results_df)
>>> report.plot_metric_bar(metric="accuracy")
>>> report.plot_metric_bar(metric="spd")
>>> tables = report.summary_tables()

__init__(results_df, metrics=None, datasets=None, methods=None, classifiers=None)

Parameters:
  • results_df (DataFrame) –

    DataFrame with columns dataset, method, classifier, plus one column per metric.

  • metrics (list of str, default: None ) –

    Explicit list of metric column names to use. Auto-detected when None.

  • datasets (list of str, default: None ) –

    Keep only these datasets. None keeps all.

  • methods (list of str, default: None ) –

    Keep only these methods. None keeps all.

  • classifiers (list of str, default: None ) –

    Keep only these classifiers. None keeps all.

plot_all(datasets=None, fairness_metric='spd', methods=None, classifiers=None)

Run all plot methods and return a list of (fig, axes).

Parameters:
  • datasets (list of str, default: None ) –

    Datasets to include. None uses all.

  • fairness_metric (str, default: 'spd' ) –

    Fairness metric for tradeoff plot (default "spd").

  • methods (list of str, default: None ) –

    Methods to include. None uses all.

  • classifiers (list of str, default: None ) –

    Classifiers to include. None uses all.
Returns:
  • list of (fig, axes) tuples –

plot_metric_bar(metric=None, datasets=None, methods=None, classifiers=None, reference_line='auto', **kw)

Grouped bar chart for a single metric across datasets.

Parameters:
  • metric (str, default: None ) –

    Metric to plot. Defaults to first performance metric.

  • datasets (list of str, default: None ) –

    Datasets to include. None uses all.

  • methods (list of str, default: None ) –

    Methods to include. None uses all.

  • classifiers (list of str, default: None ) –

    Classifiers to include. None uses all.

  • reference_line (float, "auto", or None, default: 'auto' ) –

    "auto" adds reference lines for fairness metrics.
Returns:
  • tuple of (fig, axes) –

plot_ranking(metrics=None, datasets=None, higher_is_better=None, classifier=None, methods=None, classifiers=None, **kw)

Heatmap of method rankings per dataset.

Parameters:
  • metrics (list of str, default: None ) –

    Metrics to rank on. Defaults to all detected metrics.

  • datasets (list of str, default: None ) –

    Datasets to include. None uses all.

  • higher_is_better (dict, default: None ) –

    {metric: bool} overrides for ranking direction.

  • classifier (None, "average", "best", or a classifier name., default: None ) –

    How to aggregate across classifiers before ranking.

  • methods (list of str, default: None ) –

    Filter to these methods only.

  • classifiers (list of str, default: None ) –

    Filter to these classifiers only.
Returns:
  • tuple of (fig, axes) –

plot_tradeoff(fairness_metric='spd', performance_metric='accuracy', datasets=None, methods=None, classifiers=None, **kw)

Scatter plot: |fairness| vs performance.

Parameters:
  • fairness_metric (str, default: 'spd' ) –

    Fairness metric column name (default "spd").

  • performance_metric (str, default: 'accuracy' ) –

    Performance metric column name (default "accuracy").

  • datasets (list of str, default: None ) –

    Datasets to include. None uses all.

  • methods (list of str, default: None ) –

    Methods to include. None uses all.

  • classifiers (list of str, default: None ) –

    Classifiers to include. None uses all.
Returns:
  • tuple of (fig, axes) –

summary_tables(metrics=None, datasets=None, classifier=None, methods=None, classifiers=None)

Pivot tables of metric values per method.

Parameters:
  • classifier (None, "average", "best", or a classifier name., default: None ) –

    How to aggregate across classifiers.

  • methods (list of str, default: None ) –

    Filter to these methods only.

  • classifiers (list of str, default: None ) –

    Filter to these classifiers only.

  • Returns –

to_html(path, datasets=None, methods=None, classifiers=None, metrics=None, fairness_metric='spd', performance_metric='accuracy', classifier=None)

Export an HTML report with embedded matplotlib charts.

Parameters:
  • path (str) –

    Output file path (e.g. "report.html").

  • datasets (list of str, default: None ) –

    Filters.

  • methods (list of str, default: None ) –

    Filters.

  • classifiers (list of str, default: None ) –

    Filters.

  • metrics (list of str, default: None ) –

    Metrics to include. Defaults to all detected.

  • fairness_metric (str, default: 'spd' ) –

    Fairness metric for averaged/detailed/tradeoff charts.

  • performance_metric (str, default: 'accuracy' ) –

    Performance metric for tradeoff chart.

  • classifier (None, "average", "best", or a classifier name, default: None ) –

    How to aggregate for rankings/tables.