Helps with mixers that are very expensive to run, e.g. GluonTS.
This way, we can override model prediction on a (potentially huge) validation dataset and simply report the validation loss as a proxy, for example. It's up to the provided function that will override the specified mixer within the ensemble.
Helps with mixers that are very expensive to run, e.g. GluonTS.
This way, we can override model prediction on a (potentially huge) validation dataset and simply report the validation loss as a proxy, for example. It's up to the provided function that will override the specified mixer within the ensemble.