#learning-tasks

Question 1: Single choice

Why should the learning task (classification, regression, clustering, …) be settled before a model is built?

It fixes what the model must output, which decides the training objective and the evaluation1
It fixes how many hidden layers the network must have2
It is only documentation, because a neural network works the task out from the training data by itself3
It decides which optimizer, such as SGD or Adam, training has to use4