## kernels.rolling_residuals()


Standardised forecast errors of a fixed forecaster at earlier cutoffs.


Usage

``` python
kernels.rolling_residuals(
    forecast_at,
    actual_at,
    *,
    cutoffs,
    n_periods,
    size,
    scale_floor=(0.01, 1.0)
)
```


`forecast_at(k)` and `actual_at(k)` return one value per unit, in one fixed unit order, for cutoff `k`; `horizon = n_periods - k`. Rows whose standardised error is not finite are dropped and counted in `frame.attrs["n_dropped_nonfinite"]`, as the R study filters them.
