capability__pred_int

capability__pred_int()

Capability: the forecaster can make probabilistic or interval forecasts.

• String name: "capability:pred_int"

• Public capability tag

• Values: boolean, True / False

• Example: True

• Default: False

sktime supports a range of ways to make probabilistic type forecasts, via the following methods of any forecaster:

• predict_interval: prediction intervals

• predict_quantiles: quantile forecasts

• predict_var: variance forecasts

• predict_proba: distribution forecasts

If the capability:pred_int tag is True, the forecaster can make probabilistic type forecasts using all of the above methods.

Even if the forecaster natively implements only one of the above methods, all are available to the user:

• interval and quantile forecasts are of equivalent information, with intervals of a coverage assumed symmetric

• a forecaster with available distribution forecasts obtains prediction intervals and quantile forecasts from the distribution

• a forecaster with available variance forecasts assumes a normal distribution around the predict output as mean, and derives prediction intervals and quantiles from that normal distribution

• a forecaster with available interval or quantile uses the IQR to to derive variance forecasts under normality assumptions. Users should note that this may lead to a distribution forecast which is not consistent with interval or quantile forecasts.

If the tag is False, the forecaster cannot make probabilistic forecasts, and will raise an error if a probabilistic forecast is attempted.