Calibration

polars_ts.calibration

Forecast calibration diagnostics for probabilistic forecasts. Closes #58.

calibration_table(df, actual_col='y', quantile_cols=None, quantiles=None, id_col=None)

Compute observed vs expected coverage per quantile.

Parameters

df DataFrame with actuals and quantile forecast columns. actual_col Column with actual values. quantile_cols Quantile forecast column names. Auto-detected from q_* if None. quantiles Quantile levels. Parsed from column names if None. id_col If provided, compute per group.

Returns

pl.DataFrame DataFrame with columns ["quantile", "expected_coverage", "observed_coverage"].

pit_histogram(df, actual_col='y', quantile_cols=None, quantiles=None, n_bins=10)

Compute Probability Integral Transform histogram data.

A well-calibrated model produces a uniform PIT histogram.

Parameters

df DataFrame with actuals and quantile forecast columns. actual_col Column with actual values. quantile_cols Quantile forecast column names. quantiles Quantile levels. n_bins Number of histogram bins.

Returns

pl.DataFrame DataFrame with columns ["bin_lower", "bin_upper", "count", "density"].

reliability_diagram(df, actual_col='y', quantile_cols=None, quantiles=None)

Compute data for a reliability (calibration) plot.

Parameters

df DataFrame with actuals and quantile forecast columns. actual_col Column with actual values. quantile_cols Quantile forecast column names. quantiles Quantile levels.

Returns

pl.DataFrame DataFrame with columns ["expected", "observed"] for plotting.