Coding A Neural Network FROM SCRATCH! (Part 2) 

I smell an underrated channel. You are literally the savior of my science fair project, thank you so much.

The Brain function can be heavily simplified You can put the two edge cases outside of the loop, caling layers[0], and layers[layers.length-1], and having the for loop start with i=1, and run while 1 < layers.length -1

Hey! I'm wondering when Part 3 is coming out. Can't wait to see it!

we need the continuation, really

I was following along in python. Here's the code if anyone wants it. I didn't test it though because I don't really know how to use it. Tutorial was too short ): networkShape = [2, 4, 4, 2] class Layer(object): def __init__(self, n_inputs, n_nodes): self.n_nodes = n_nodes self.n_inputs = n_inputs self.weightsArray = [n_nodes, n_inputs] self.biasesArray = [n_nodes] self.nodeArray = [n_nodes] def forward(self, inputsArray): self.nodeArray = [self.n_nodes] for i in range(self.n_nodes): # Sum of the weights times inputs for j in range(self.n_inputs): self.nodeArray[i] += self.weightsArray[i, j] * inputsArray # Add the bias self.nodesArray[i] += self.biasesArray[i] def activation(self): for i in range(self.n_nodes): if self.nodeArray[i] < 0: self.nodeArray[i] = 0 def awake(): global layers # layers = Layer(len(networkShape) - 1) layers = [] for i in range(len(networkShape) - 1): # layers[i] = Layer(networkShape[i], networkShape[i + 1]) layers.append(Layer(networkShape[i], networkShape[i + 1])) def brain(inputs): for i in range(len(layers)): if i == 0: layers[i].forward(inputs) layers[i].activation() elif i == len(layers) - 1: layers[i].forward(layers[i - 1].nodeArray) else: layers[i].forward(layers[i - 1].nodeArray) layers[i].activation() return layers[-1].nodeArray

I've seen a lot of videos about Neural Networks and yours is the one that explains it in an understandable manner (Or maybe the 10th time is the charm) I'm curious to see the next one

Great video. Thanks for the effort. I'm looking forward to see the part 3. Cheers,

A great series. Not only for content, but well edited too. Cheers John.

Let's gooooo, part 3 please!

Nice video! In general I'd say a neural network is still a black box even if you built it and know the values of all the nodes, weights, biases and layers.

Hopefully you'll finish this eventually! I enjoyed the last two videos.

isn't the 'layer' in layer[ i ] = new Layer( networkShape[ i ], networkShape[ i + 1 ] ); supposed to be 'layers' ?

Hey! I wonder when episode 3 will come out?

Great video ! I'm really looking forward to see how the network will be trained

That was an excellent, practical video on neural networks! As someone just beginning to dig into this subject, I love it! Also, clean and neat code. Enjoyable to read (although I'm not a fan of nesting classes).

Please Upload Part 3

thank you so much! this is exactly what i need for my uni project

just amazing 🤩🤩

part 3 how to set up in unity?

best video ive ever seen not gonna lie

amazing

Hi, is part 3 coming out?

drop the training video right now!

amazing !!! waiting the training part

O man thank you! Such a gem content out here :) I'm a Swift Dev the code is not hard to grasp

Part 3?

I wonder that the shape of network . I mean how many hidden layers and nodes should we use for each sample . And also wondering about third part .

very useful John

:') Episode 3 where are you... this is the new Half Life 3 for me. Am I wrong in thinking the weights and biases were never given values here? should those be made in this class too?

this is a much better way to do the forward pass: public void Forward(float[] inputsArray) { for (int i = 0; i < n_nodes; i++) { nodeArray[i] = biasesArray[i]; for (int j = 0; j < n_inputs; j++) { nodeArray[i] += weightsArray[i, j] * inputsArray[j]; } } } this way you don't create a new array every time. also the opening "{" is in the correct location.

please, do the third part

very good :)

i need that next video i have no idea what im doing :( i have this code and i think i understood how it works after starring at it for an Eternity BUT how can i make use of it now....

if the activation function always happen right after forward. why not just combine them?

Thank you! I would have kissed you for that great explanation!

Good tutorial, but as others have pointed out there is a compile error both in the video and the github code in Awake(). layer in the for loop should be layers. Makes me wonder if was ever tested?

what

This is useless, you literally just implemented matrix dot in C#. Most of the difficulty in making a neural network is just backprop, jfc