If I use the dataset transformed by pycaret and the best trained model why don't I get the same results I get with pycaret? I am trying to learn how to build regression models with datasets downloaded from the internet (boston dataset, iris, etc.) but with pycaret I get good results, if I write the program from scratch I don't.
First in pycaret I ran the setup() command. Then after choosing the best model from compare_models() and finalize it with finalize_model() I download the dataset initially transformed by pycaret. At this point if I use the transformed dataset and the tuned model parameters obtained from print(tuned_model) shouldn't I be able to do it myself? Of course as pycaret I divide the dataset into train, test and validation set. Which step am I skipping?
