Auto news on Youtube Sep 2 2018

What is up guys, it's Adam back with another video and today we will be building a simple neural network in Python

But before we get started

I'd like to give a shout out to at I am Trask for writing the code used in this video

I'll give a link in the description to his blog post on neural nets without further ado. Let's get started

Alright guys, so I am currently an Adam text editor and we are going to start our Python file by importing numpy as MP

Next we're going to define our sigmoid function which in short if you don't understand what that is

It'll basically make all of our weights into one scale that goes from 0 to 1

all we are really doing is implementing the math that you see on screen right now, so

This function will have two parameters X and derivative

which will be set by default as false now if

Drivel to true we are going to return x times 1

Minus X and then outside of that if statement we will return 1 over 1 plus NP

exp for exponent 2 negative X

Now that our sigmoid function is defined. Let's write in the X and y components

For our neural net to train on so first we will set X as equal to NP or a open brackets

0 0 1 0

1 1

101 and

finally 1 1 1 so this is just going to be our input data and

now for the corresponding Y values y equals NPR a

0 0 1 1 dot T capital T that is for transpose so that we can use matrix

multiplication

So just to recap our neural net is trying to predict what the value of y is based on X

So if X is equal to 0 0 1 as it is in the first training sample

the output should be 0 because that is what the course why value for the x-component is

Next we are going to initialize a seed value with NP random

This will mean that the neural net will initialize the same random value every time

This is a good practice to always use and it will help a lot with debugging

Now we are going to make our synapse which if you know anything about neurons in our brain

It's something that connects two neurons to each other since we only have one hidden layer

We are only connecting the input to the output of our network

the input being our three x values and the output being our 1 Y value so

We will set a variable called sin 0 equal to 2 times MP dot random

3 comma 1

minus 1

This is for the 3 input X values and the 1 Y value

All right

so finally we are going to start training our network to do this start a for loop and iterate through it as many times as

You like a good number to start with for now will be?

10,000 if

This number is too low the network won't be able to update the weights enough for them to make the neural network work

Well, if it is too high, we will over fit our data making the neural net not work. Well with new data

Ok, so in our for loop layer 0 or the input layer is set to X and layer 1 or the output layer is set

to the sigmoid of NP dot dot which is for matrix multiplication l 0 sin 0

So basically we are multiplying the input layer by our synapse and setting it in a value from 0 to 1

Now, let's set a variable called L

1 underscore error

Which will compare our L 1 value to the actual Y value as we iterate through our loop this variable should get smaller and smaller

As our neural net gets more accurate

Next let's set a variable called l1 Delta to the l1 error times the sigmoid of l1

With the deriv equals true so that it runs that if statement we defined earlier

This will help all neural net not become too overconfident in its answers

Now, let's make synapse zero two plus equal NP

dot layer zero dot transpose and l1 Delta

This line will update the weights of our network

Finally outside of our for loop. Let's print out layer one

So our neural net is now basically done. Let's run it and see what the results are

You can see our network is pretty accurate

It thinks the first two X values are probably zero and then the last two are probably one

Congratulations, you just built your first neural network

So if you didn't understand anything that's going on in this code. That's okay

There's so much theory behind a lot of these concepts and it takes time to fully

Understand everything if you want to continue working with neural nets and machine learning

Please comment down below and I will also do a video on how I'm learning machine learning personally

Also, once again, thank you to at ion traps for writing the original code for this video

I think it's a really good introduction to neural nets. You can check them out below