Glossary

This glossary defines key terms and concepts used throughout the k-diagram documentation and package.

ACov (Angular Coverage)

A parameter (acov) controlling the angular span of the polar plots, such as ‘default’ (360°), ‘half_circle’ (180°), ‘quarter_circle’ (90°), or ‘eighth_circle’ (45°).

Anomaly (Prediction)

An instance where the observed (actual) value falls outside the predicted uncertainty interval (i.e., below Qlow or above Qup). Visualized by plot_anomaly_magnitude().

Anomaly Magnitude

The severity of a prediction anomaly, measured as the absolute distance between the actual value and the nearest violated prediction interval bound.

Calibration (Interval)

The degree to which the empirical coverage rate of prediction intervals matches their nominal coverage rate. A well-calibrated 90% interval should cover approximately 90% of the actual values.

Consistency (Interval Width)

The stability or variability of the prediction interval width (Qup - Qlow) for a specific location or sample across multiple time steps or forecast horizons. Assessed by plot_interval_consistency().

Coverage (Empirical)

The actual fraction or percentage of observed (true) values that fall within their corresponding prediction intervals in a given dataset. Calculated by plot_coverage() and visualized point-wise by plot_coverage_diagnostic().

Coverage (Nominal)

The theoretical or intended coverage rate of a prediction interval, determined by the quantile levels used. For example, the interval between the 10th (Q10) and 90th (Q90) percentiles has a nominal coverage of 80%.

Drift (Model / Concept)

The degradation or change in a model’s performance or underlying data relationships over time or changing conditions.

Drift (Uncertainty)

The change in the magnitude or pattern of predicted uncertainty (typically interval width) over time or forecast horizons. Visualized by plot_model_drift() (average drift) and plot_uncertainty_drift() (pattern drift).

Fingerprint (Feature)

A characteristic profile of feature importance values for a specific model, time period, or group, often visualized using a radar chart via plot_feature_fingerprint().

Interval Width

The difference between the upper quantile (Qup) and lower quantile (Qlow) of a prediction interval, representing the magnitude of predicted uncertainty. Visualized by plot_interval_width().

K-Diagram

The term used for the specialized polar diagnostic plots generated by this package, named after the author (Kouadio).

Polar Plot / Coordinates

A graphical system where points are located by an angle (theta, θ) and a distance from a central point (radius, r). Used extensively in k-diagram.

Prediction Interval (PI)

A range [Qlow, Qup] derived from quantile forecasts, intended to contain the actual observed value with a certain probability (nominal coverage).

Quantile

A value below which a certain proportion of the data or probability distribution falls. Common examples used in forecasting are Q10 (10th percentile), Q50 (50th percentile or median), and Q90 (90th percentile).

Radar Chart

A type of polar plot where multiple quantitative variables (represented by axes radiating from the center) are shown for one or more observations (represented by polygons or lines). Used by plot_feature_fingerprint() and optionally by plot_coverage().

RMSD (Centered Root Mean Square Difference)

A metric implicitly represented on a Taylor Diagram as the distance between a model point and the reference point. It measures the overall difference considering both standard deviation and correlation.

Taylor Diagram

A polar-style diagram summarizing model skill by plotting standard deviation (radius), correlation (angle), and RMSD (distance from reference) relative to observed data. Generated by functions in kdiagram.plot.evaluation.

Uncertainty Quantification (UQ)

The process of estimating and characterizing the uncertainty associated with model predictions, simulations, or measurements.

Velocity (Prediction)

The average rate of change of the central prediction estimate (e.g., Q50) over consecutive time steps for a given location or sample. Visualized by plot_velocity().