If we look at the lesson 4 notebook Tabular data,
It has cat_names and cont_names
For my dataset, it has a user id. I am thinking that I should not pass user id as a variable for the neural network model because user id does not affect the prediction. Or I can still pass it because the model will try to forget it?
If you know there is no relation with prediction and a column, not including it will make it easier for the model to learn. Any field randomly generated that appears only once in every row could be ignored. I wouldn’t say the same for an auto increment id one, as that can be an indirect indicator of how long has the record existed.
Your intuition is correct - if the variable does not affect the prediction (user id in your case), then you should not pass it to the nnet model. I don’t think there is any reason to increase the dimensions of the training dataset by adding another column which does not help improve its performance.
Even in cases when we have a lot of features to train a model, we tend to do feature selection to use only the features which help improve the model performance while reducing the training time, so I don’t see any reason to add a feature which we know for sure does not helps.