As discussed on Monday, cross-validation gives an error estimate for a model trained on all available data. Therefore, the all-data-trained model is used as the final model, not one of the models created in the cross-validation routine.
This figure from the Google crash course suggests using the best model from within the cross-validation routine, which is inconsistent.
As discussed on Monday, cross-validation gives an error estimate for a model trained on all available data. Therefore, the all-data-trained model is used as the final model, not one of the models created in the cross-validation routine.
This figure from the Google crash course suggests using the best model from within the cross-validation routine, which is inconsistent.
https://github.com/geco-bern/agds/blob/7c71dd3a9a9d6aac2c684a3eeec0967a071d3a46/10-supervised_ml_II.Rmd#L199