Skip to content

check_interval_bounds

yohou.testing.interval.check_interval_bounds(forecaster, y_test, X_actual_test=None)

Check upper >= lower for all coverage rates and time steps.

Parameters

Name Type Description Default
forecaster BaseIntervalForecaster

Fitted interval forecaster instance

 required
y_test DataFrame

Test target data

 required
X_actual_test DataFrame

Test features

 None

Raises

Type Description
AssertionError

If upper bounds are less than lower bounds

Source Code

View on GitHub

Show/Hide source 
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
def check_interval_bounds(forecaster, y_test: pl.DataFrame, X_actual_test: pl.DataFrame | None = None) -> None:
    """Check upper >= lower for all coverage rates and time steps.

    Parameters
    ----------
    forecaster : BaseIntervalForecaster
        Fitted interval forecaster instance
    y_test : pl.DataFrame
        Test target data
    X_actual_test : pl.DataFrame, optional
        Test features

    Raises
    ------
    AssertionError
        If upper bounds are less than lower bounds

    """
    forecasting_horizon = min(3, len(y_test))
    y_pred = forecaster.predict_interval(forecasting_horizon=forecasting_horizon)

    coverage_rates = forecaster.fit_coverage_rates_

    # Check if we have panel data (columns with __ separator)
    _, y_panel_groups = inspect_panel(y_test)

    if len(y_panel_groups) > 0:
        # For panel data, interval columns use __ separator
        for group_prefix in y_panel_groups:
            # Get fields from the original training data (full column names)
            expected_fields = y_panel_groups[group_prefix]

            for rate in coverage_rates:
                for field in expected_fields:
                    lower_col = f"{field}_lower_{rate}"
                    upper_col = f"{field}_upper_{rate}"

                    lower_vals = y_pred[lower_col].to_numpy()
                    upper_vals = y_pred[upper_col].to_numpy()

                    violations = lower_vals > upper_vals
                    if violations.any():
                        raise AssertionError(
                            f"Found {violations.sum()} violations where lower > upper for {field} at coverage {rate}"
                        )
    else:
        # For global data, check individual columns
        target_cols = list(forecaster.local_y_schema_.keys())

        for rate in coverage_rates:
            for col in target_cols:
                lower_col = f"{col}_lower_{rate}"
                upper_col = f"{col}_upper_{rate}"

                lower_vals = y_pred[lower_col].to_numpy()
                upper_vals = y_pred[upper_col].to_numpy()

                violations = lower_vals > upper_vals
                if violations.any():
                    raise AssertionError(
                        f"Found {violations.sum()} violations where lower > upper for {col} at coverage {rate}"
                    )