Combination of train and validation set (and even test set)

[moto]

Dear ADDI team,

I am wondering if it is fine to combine the train set and validation set when modeling ? I just wanted to use all available labels.

In term of feature engineering, it is accepted to combine the train and the test set in order to impute the missing values … ?

Many thanks.

[michael_bordeleau]

It’s common practice in machine learning that you can use all the data you can your hands, but that comes at a risk of overfitting.

Hopefully team says likewise.

Ah, and you just spoiled early the trick of combining train+valid.