Tensorflow: Tutorial on "How to build a reliable stack of layers"

Dear Freecodecamp and coders,

I ran thru the education-plan several times but still have no clue how to build a reliable stack of layers.

Can someone provide me with one or more links with tutorials to this issue?

Thanks :slight_smile:

What do you mean with “reliable”?
If we could calculate on how to make a fitting set of layers, we could automate that.

I know this might not be the most satisfying answer. But it’s a lot of trial and error or requires some very advanced knowledge.
You can try this: Introduction to TensorFlow  |  Machine Learning Crash Course

But I don’t think you’ll find a lot of free resources on the topic.

100 percent “try and error” is not reliable to me. I right now passed the second ML test with a bit of help and hopefully I will find a logic to create pattern which are not “try and error”.

Let me emphasize the “advanced knowledge” part: I would think that it’s something that basically goes beyond the understanding achieveable in short online sources.
There are books and courses and propably entire university degrees dealing with machine-learning and neural-networks.

Unfortunately, the problems of machine learning is essentially as complex as human cognition and learning. Put another way, if we can learn how to design perfect neural networks to do a task, say classify cats or dogs, then we’ll likely have insight into how humans actually make that classification, and vice-versa.

Think about training yourself as a cat or dog classifier. You had to have someone who told you whether the animal in question was a cat or dog, and while you probably learned quickly, it was by trial and error just like training a neural network. Our neural network is the product of eons of trial and error; we just call it natural selection and our network is very good.

For more information, you really need university level courses or textbooks (here’s a popular free one). You can thumb through some of the less mathematical chapters and get an idea of what kind of models work for solving certain problems but I’m afraid that at the end of the day the specialists are still only really good at trial and error.

There’s also the possibility that ML turns out like the second law of thermodynamics and trial and error is as good as it gets because that is the answer.