The question concerns the study of neural networks in Python. Moreover, for a person who has no idea what this network is and how it works. Google did not give any useful links, any sane articles, where the Russian language for an uninformed person explained what it is and what it is eaten with. I, of course, exaggerate a little, and after a generalized excursion into biology, one can imagine how neural networks work in general, but here it begins: “Neural networks have a nonconvex probability function, which leads to the problem of local maxima”, “Initializing the weights matrix with random data "," Let's create a network of 10 neurons, 20 layers and 100500 inputs "," Let's give the input signal and teach the network to distinguish dark beer from light ones, "etc. etc. Nobody anywhere explains why for a particular task, how many neurons are needed, two, not eighty layers, what kind of data and in what form do they go to the input? How many kilograms weigh on scales? Is this a mystery or am I completely stupid? Explain ...

1 answer 1

No one anywhere explains why, for a specific task, it is exactly how many neurons are required, two, not eighty layers.

This is one of the tasks that a person is currently solving using the selection method based on his experience in designing neural networks. (Another serious challenge is the way of learning)

On neural networks I can recommend a great introduction .

As for Python, to begin with, PyBrain is enough for you to play - this is, for the time being, I think, the simplest library.

    Protected by a member PashaPash ♦ May 27 '18 at 22:15 .

    Thank you for your interest in this issue. Since he collected a large number of low-quality and spam responses, which had to be deleted, now it’s necessary to have 10 reputation points on the site (the bonus for account association is not counted ).

    Maybe you want to answer one of the unanswered questions ?