A reply to `Do we need use full dataset to train a model with the hyper parameters chosen by cross validation`?

I wonder if we should use full dataset to train model with the hyper parameters that have the best performance in cross validation? Since in the cross validation, we usually use part of the dataset to train and remain for test, which means that the best parameters in cross validation may not be the finest for the model trained on full data.

And then I find some good reply for this question.

• https://machinelearningmastery.com/train-final-machine-learning-model/
• https://stats.stackexchange.com/questions/52274/how-to-choose-a-predictive-model-after-k-fold-cross-validation
• https://stats.stackexchange.com/questions/11602/training-on-the-full-dataset-after-cross-validation