What dataset in the final deployment we should use?

Course Q&A Machine Learning Specialization Supervised ML: Regression and Classification
,
1

Hi All,

In machine learning specializaiton, I am on Advance Algorithms / Week 3 –Advice for applying machine learning.

Using cross validation is a nice trick to pick a model, my question is about deployment stage.

What is the better practice in deployment: (1) use train set, check error/accuracy on cross validation set to pick a model, and then use test set to report train, and immedately deploy this to production or (2) after train-validation-test done and retrain the picked model for whole dataset including training/crossvalidaiton/test and deploy the new parameters to production?

Thanks,

2

Never use the test set in training. It’s your independent verification that the model works well enough to meet your goals.