I have studied a dataset and built a model. During the exploration I made several transformations and before training the model I applied a pandas pipe to my train dataset. This pipe includes transformations such as setting ‘unknown’ values to nan or binarize some feature, but also includes dropping outliers.

My question is how to manage those changes once I want to predict new data with my model. I know I should do the same transformations so that the process make sense, but I can not drop values I am supposed to predict even if they are outliers. It is correct to create another pandas pipe with only some of the steps from the original pipe and apply it to new data to predict?