You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In a scenario where we want to fuse the output of several models, it could be interesting to be able to train models on the output of previous model predictions.
Note that this also implies some specific model selection strategies in order to train the fusion model on data sets which were not used to train the previous models.
Discussed till now:
Use sepearate prediction jobs for previous forecasts, these aim at forecasting the same target but use different models. These pj's must have validation data that match
Use another prediciton job to combine above predictions using a quantile method. To train we need historic predictions. Can we use vaildation data?
Use a back test for this? Have a option to save the backtest model.
The text was updated successfully, but these errors were encountered:
In a scenario where we want to fuse the output of several models, it could be interesting to be able to train models on the output of previous model predictions.
Note that this also implies some specific model selection strategies in order to train the fusion model on data sets which were not used to train the previous models.
Discussed till now:
Use sepearate prediction jobs for previous forecasts, these aim at forecasting the same target but use different models. These pj's must have validation data that match
Use another prediciton job to combine above predictions using a quantile method. To train we need historic predictions. Can we use vaildation data?
Use a back test for this? Have a option to save the backtest model.
The text was updated successfully, but these errors were encountered: