BoostAODE: make the predict in boosting with the ensemble instead of the last model #19

New Issue

rmontanana · 2024-02-26T12:50:56Z

rmontanana commented

2024-02-26 12:50:56 +00:00

Probabilities table

We shall have a probabilities table that is going to be update with each new model that is added to the ensemble, this way when we use this table to compute predict, we are going to take into account each model of the ensemble for the prediction.

Feature selection

The models selected in the initialization process all have to be trained with 1/m weights but shall update the Probabilities table. At the end of the initialization process (when the selected features are over) we'll use the weights obtained with the last predict (that predict has been computed taking into account all the features selected) and with the computed alpha_t we'll update the significance of the models computed in the initialization process (all these models will have the same significance)

## Probabilities table We shall have a probabilities table that is going to be update with each new model that is added to the ensemble, this way when we use this table to compute _predict_, we are going to take into account each model of the ensemble for the prediction. ## Feature selection The models selected in the initialization process all have to be trained with 1/m weights but shall update the _Probabilities table_. At the end of the initialization process (when the selected features are over) we'll use the weights obtained with the last predict (that predict has been computed taking into account all the features selected) and with the computed alpha_t we'll update the significance of the models computed in the initialization process (all these models will have the same significance)

rmontanana added this to the v1.0.4 milestone 2024-02-26 12:50:56 +00:00

rmontanana added the

💡 enhancement

label 2024-02-26 12:50:56 +00:00

rmontanana self-assigned this 2024-02-26 12:57:17 +00:00

rmontanana referenced this issue from a commit

2024-03-06 16:16:18 +00:00

Merge pull request 'predict_single' (#22) from predict_single into main

rmontanana closed this issue

2024-03-06 16:16:20 +00:00

Sign in to join this conversation.