I originally posted this is SMP, and was redirected over here.

Thanks for any direction you can provide!

Well, it is doable from first principles. I did my very first neural network stuff in matlab (netural networks at a basic level can be modeled by matrix operations quite well). I wrote some stuff in C/C++ after that. I wrote some genetic programming stuff in C++ and wrote some genetic algorithm stuff in Tcl (which has some lisp like properties and is kind of a good choice for it, imo). I worked on some other techniques also, such as simulated annealing. I did my graduate research in using these techniques to solve some real world problems.

That said, the real problem with the field, at least when I was working in it, is that sort of the "toy" stuff that you see in textbooks is pretty easy to do and understand, and the more real your problem gets, the harder it is to characterize. Take neural networks for example, you have a lot of choices for feedback methods, size of layers and number of layers, and some other stuff. Your chosen problem might really only be solvable with certain magic parameters and a lot of your research turns into either selecting these parameters, or writing code in a way that throws everything at the wall until it sticks (trying a ton of techniques to avoid local minimums and converge faster on a solution)

And soooo many times you think you've nailed it and find you've over-trained, or that your solution isn't stable over time, etc, etc. It's very frustrating.

(but also a very interesting field)

There's a video series I just started watching about Deep Learning using Python: https://www.youtube.com/watch?v=oYbVFhK_olY.

Deep Learning is technically a neural network with more than 1 hidden layer.

I'm okay with this. I'm just playing around with it because I think it's interesting. I have no grand aspirations of getting into the field and doing some sort of meaningful research.

I appreciate the link. I'll have a look at it over the weekend.

I'm sure you have the math skills to begin with something that doesn't hold your hand as much as this does, but I (being bad at math) found it useful.

I just want to add that I found out he also has another series on machine learning that covers both the theory and the libraries, but also spends time building up the calculations from scratch.

https://pythonprogramming.net/machin...-introduction/

Yea, actually sentdex has something like 800+ videos on Youtube... pretty amazing and covers a wide range of topics.