Do you merge your validation and training sets?

Question

I'm currently implementing a K-Nearest Neighbours model and I'm at the stage of splitting up the datasets for cross-validation.

I understand the need for the Training, Validation and Test sets, though one thing I'm unsure about is what to do with the Validation and Test sets after the model has been tuned and tested.

Is there any reason for not merging them back into the training set to provide more data? Or do I now leave them and refrain from touching them again?

Answer 1

As far as I know, you want to use all the data for the actual deployment of your machinery. Definitely do what you say but when the time has come, you want to train your machines with the most data you can get your hands on.