ConfidenceRegionDisplay#

class mlquantify.visualization.ConfidenceRegionDisplay(prev_estims, *, confidence_level=0.95, class_names=None, true_prevalence=None)[source]#

Confidence region around a single prevalence prediction.

Visualises the uncertainty of one quantifier prediction, obtained from a set of bootstrap prevalence estimates (see mlquantify.confidence). The layout adapts to the number of classes:

3 classes (kind='ternary'): the bootstrap estimates are projected onto the 2-simplex and a confidence ellipse is drawn, together with the point estimate and (optionally) the true prevalence.
otherwise (kind='interval'): a per-class “forest” plot of the point estimate with a percentile confidence interval per class.

This is a single-sample display.

Parameters:

prev_estimsndarray of shape (m, n_classes): m bootstrap prevalence estimates for one prediction.
confidence_levelfloat, default=0.95: Confidence level of the region.
class_nameslist of str, default=None: Class labels in column order.
true_prevalencearray-like of shape (n_classes,), default=None: Optional ground-truth prevalence to overlay.

Attributes:

point_estimate_ndarray of shape (n_classes,): Mean of the bootstrap estimates.
ellipse_matplotlib Ellipse or None: The confidence ellipse (ternary layout only).
scatter_matplotlib PathCollection or None: The projected bootstrap cloud (ternary layout only).
errorbar_matplotlib ErrorbarContainer or None: The per-class intervals (interval layout only).
ax_matplotlib Axes
figure_matplotlib Figure

See also

mlquantify.confidence.construct_confidence_region: Builds the regions.
PrevalenceDisplay: Plain per-class prevalence bars.

Examples

>>> import numpy as np
>>> from mlquantify.visualization import ConfidenceRegionDisplay
>>> estims = np.random.dirichlet([8, 6, 6], size=300)
>>> disp = ConfidenceRegionDisplay.from_estimates(   
...     estims, confidence_level=0.95, class_names=["a", "b", "c"])

classmethod from_estimates(prev_estims, *, confidence_level=0.95, class_names=None, true_prevalence=None, ax=None, **kwargs)[source]#: Build a display directly from bootstrap prevalence estimates.

classmethod from_region(region, *, class_names=None, true_prevalence=None, ax=None, **kwargs)[source]#

Build a display from a fitted mlquantify.confidence region.

Parameters:

regionBaseConfidenceRegion: A region instance exposing prev_estims and confidence_level (e.g. the output of mlquantify.confidence.construct_confidence_region).
class_nameslist of str, default=None
true_prevalencearray-like, default=None
axmatplotlib Axes, default=None
**kwargs: Passed to plot.

Returns:

displayConfidenceRegionDisplay

plot(ax=None, *, kind='auto', name=None, **kwargs)[source]#

Draw the confidence region.

Parameters:

axmatplotlib Axes, default=None: Axes to draw on.
kind{‘auto’, ‘ternary’, ‘interval’}, default=’auto’: Layout. 'auto' selects 'ternary' for 3-class problems and 'interval' otherwise.
namestr, default=None: Label for the point estimate.
**kwargs: Forwarded to the primary artist (ax.scatter of the bootstrap cloud for ternary; ax.errorbar for the interval layout).

Returns:

displayConfidenceRegionDisplay