Skip to main content

​
make_interval_method

Adapt a ConformalMethod to mlforecast’s interval-hook signature.

​
Parameters

method_cls : type[ConformalMethod] A method class, e.g. :class:~conformalforecast.methods.split.SplitConformal. alpha is supplied per requested level by the returned function, so do not pass it here. **method_kwargs Extra keyword arguments forwarded to method_cls (gamma, rho, seed, score, …). A score must be an :class:AbsoluteResidual, like the class default it replaces.

​
Returns

callable A function matching mlforecast 1.1’s interval hook — fcst_df and cs_df positional, then keyword model_names, level, cs_n_windows, cs_h, n_series, horizon, weights, is_transfer, target_weights — that appends {model}-lo-{level} / {model}-hi-{level} columns. One method is constructed and fitted per (model, level) pair. Weighted and transfer conformal arguments raise NotImplementedError.

​
Raises

ValueError If method_cls CONSUMES_QUANTILES (it needs quantile forecasts mlforecast does not record), or the score it would fit with is not an :class:AbsoluteResidual (see the module docstring).

​
Examples

from conformalforecast.compat.mlforecast_shim import make_interval_method from conformalforecast.methods import AdaptiveConformalInference hook = make_interval_method(AdaptiveConformalInference, gamma=0.01) # doctest: +SKIP

​
register_interval_method

Register a method-backed interval method in mlforecast’s registry.

​
Parameters

name : str Registry key, e.g. "conformal_method_split". Note the upstream gap described in the module docstring: PredictionIntervals validates method against a hard-coded list, so a brand-new name is reachable via get_conformal_method(name) but not yet via PredictionIntervals(method=name). method_cls : type[ConformalMethod] Method class; see :func:make_interval_method. overwrite : bool, default False Allow replacing an existing registry entry (e.g. shadowing "conformal_error" to compare against mlforecast’s native method). **method_kwargs Forwarded to method_cls.

​
Returns

callable The registered hook.

​
Raises

ImportError If mlforecast is not installed. ValueError If name is already registered and overwrite is False, or if method_cls is unsupported (see :func:make_interval_method).