any of you are comfortable and/or capable of creating a neural network from scratch?

Question

I have an idea but none of my peers and i are capable of creating a self learning neral network.

Answer 1

I have an indeed created a neural network in Python & Lua from scratch. In Python I did use some number libraries, although nothing that would play a major role. The number libraries I used were simply to help me out with small things such as random numbers (numpy) for weights & bias.

I didn't create a very large neural network (it had 0 hidden layers) although it took any number of inputs and then gave an output. I tested it with determining the color of a flower based on petal size, the score a student would get based on previous scores and time studying, and its accuracy was 98%. Essentially, you input training data in and using that data it trains its weights & bias over 50,000 times to get an extremely high accuracy. If you want help, feel free to let me know :)

EDIT: I suggest you learn very basic calculus & algebra if you haven't already. They're essential in creating a NN and without them you won't get very far.

Answer 2

You mean like, totally from scratch? There are at least three good and recommendable libraries, which are heavily optimized to do that. Don't get me wrong, but there is a rather little chance you'd make something better than those ;)

Check out keras with tensorflow back-end. It's like building with Lego blocks!

By the way, what problem are you trying to solve with NNs?

Answer 3

Thank you! I will try and check them out. Traffic congestion :)