Is nonnegative hierarchical time series reconciliation implemented in sktime? #4645
-
I'm using sktime to perform time series forecasting and hierarchical reconciliation. For the latter I am using the I am aware of other implementations of this restriction for hierarchical TS reconciliation, like the library |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
hm, interesting question. The The easiest way to get nonnegative forecasts are using the This is not exactly the same as you describe, as this is after applying the reconciler, and the nixtla forecaster has this as a parameter of the method. We're currently in the process of interfacing the nice nixtla libraries under consistent interfaces: A de-novo implementation of course would also be useful, if it can be easily added by using the internal methods of the |
Beta Was this translation helpful? Give feedback.
hm, interesting question. The
Reconciler
orReconcilerForecaster
themselves do not have this.The easiest way to get nonnegative forecasts are using the
FunctionTransformer
with the functionlambda x: max(0, x)
or similar, and using it in a pipeline after theReconcilerForecaster
, i.e.,my_reconciler * FunctionTransformer(nonneg_fun)
This is not exactly the same as you describe, as this is after applying the reconciler, and the nixtla forecaster has this as a parameter of the method.
We're currently in the process of interfacing the nice nixtla libraries under consistent interfaces:
#4643
so, if you'd like to help by contributing an interface to the hierarchical forecasters (e.g.,
NixtlaR…