Assuming that the same features are used for hyperparameter tuning and model fitting, then does the order of columns used during the hyperparameter tuning process should be identical to the order of columns used when fitting the model based on the best hyperparameters?
I am using sklearn’s make_column_transformer functions for the hyperparameter tuning CV pipeline. However, I found that this function modifies the order of provided columns when setting the remainder argument to ‘passthrough’. Should I ensure that when fitting the final model the same order of columns is preserved, or the order does not matter as long as I am using the same features.