Retrain on whole Dataset?


This is a bit more general ML, so I apologize in advance if this isn’t the right place for such a question:

If I am training a model, standard practice would be to split the data sets into 2 or 3 sets, and hold one out to assess the model accuracy (or whatever performance metric), and another one for hyperparameter tuning.

When I am done, happy with all hyperparameters and models, should I re-fit the model on the entire dataset before releasing it? Wouldn’t I want the final, final model to have the biggest dataset possible, for inference on truly new data?