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
{{ message }}
This repository has been archived by the owner on Jun 22, 2024. It is now read-only.
An aggregation module to ensemble the forecasts of various models by simple or weighted averaging.
Describe the solution you'd like
The inputs are the forecasting steps of different models. The module is initialized with the parameters weights, k_best, and loss (all optional). The loss is a list of error metrics from the given forecasting models (equal sequence!). The error metrics can be given as StepInformation (i.e., calculated with the MaeCalculator or RmseCalculator) or plain float numbers.
Possible configurations for weights and k_best:
weights = None, k_best = None -> averaging (default)
weights = None, k_best = 'auto' -> averaging k-best with k based on loss values
weights = None, k_best = k -> averaging k-best with given k
weights = [0.3,0.7], k_best = None -> weighting based on given weights
weights = [0.3,0.7], k_best = 'auto' -> weighting based on given weights and k based on loss values
weights = [0.3,0.7], k_best = k -> weighting based on given weights and k
weights = 'auto', k_best = None -> weighting with weights based on loss values
weights = 'auto', k_best = 'auto' -> weighting k-best with weights and k based on loss values
weights = 'auto', k_best = k -> weighting k-best with weights based on loss values and given k
Automated weight estimation and evaluation of k-best are only performed in pipeline.train(). Weights and chosen models are retained for pipeline.test().
The text was updated successfully, but these errors were encountered:
What is your feature request related to?
Please describe the module/problem
An aggregation module to ensemble the forecasts of various models by simple or weighted averaging.
Describe the solution you'd like
The inputs are the forecasting steps of different models. The module is initialized with the parameters
weights
,k_best
, andloss
(all optional). The loss is a list of error metrics from the given forecasting models (equal sequence!). The error metrics can be given as StepInformation (i.e., calculated with the MaeCalculator or RmseCalculator) or plain float numbers.Possible configurations for
weights
andk_best
:Automated weight estimation and evaluation of k-best are only performed in
pipeline.train()
. Weights and chosen models are retained forpipeline.test()
.The text was updated successfully, but these errors were encountered: