What are the types of goals when training a model in Keras?

Question

Neural networks

The Keras model has a fit method that is passed samples and targets. From the examples it is not clear what form the goals can take. I only realized that it can be a number for classification or regression, or a binary matrix for categories, obtained, for example, by the to_categorical function. Or maybe a goal, for example, a sequence of words? For example, 1,2,3 (1:"I", 2:"I love", 3:"Cheese", etc.). What should a numpy array look like in this case? What other types of goals can there be?

Answer 1

Target is something from which you can take a derivative, that is, only numbers.
In the case of words, one usually predicts the probability of each of the words using softmax.