Now we made this game.

Asteroids in another tutorial, Siri's so feel free to check that out and download the code from the links below.

There should be Sounds folder and an HTML file, so that's up in the game up.

So there's a triangular ship that can be moved around the string, and it can shoot yesterday.

I'll just meet that.

So the goal of this tutorial, Siri's, is a very simple one, and intentionally so because new neural networks can be quite challenging.

The goal is to rotate towards the closest asteroid and shoot, rotate to the next closest asteroid and shoot.

It's not gonna be an algorithm.

We're going to train and Neural Network to do this.

So what is a neural network?

Well, here it is.

Basically, they're made up of a number of layers, so in the import layer, a hidden layer and an output layer, you can't have multiple hidden layers, But we'll focus on just having a single hidden layer to keep things simple.

Now, the input layer it will take out game information so we could pass things such as the angle to the nearest asteroid or the current angle of the ship.

The hidden layers more like a processing layer, I guess.

And the output?

Well, we'll probably have a single output whether to turn left or turn right.

So that'll be a single Bullen.

So zero will mean left.

One will mean right.

We'll keep things nice and simple because these things can be quite complicated.

In between each of these layers is a whole bunch of black errors.

Now each of these black errors represent, await, await between minus one and one.

I believe so.

For each neuron in the previous layer, it sends out information to each of the neurons in the next layer.

So ultimately will be sending in some training data for example, the asteroid position and the ship's current angle or something similar.

And we'll feed that through with unexpected output, so either turn left or turn right now.

If it gets it wrong, will send that information back to the be in port, I guess, and use that to correct these whites so ultimately will be sending in thousands of points of information.

Correcting these waits until finally, we have a unique set of whites that can take any of our gain information and give us an accurate and reliable output.

So here's the position of an asteroid.

What do I do?

Turn left.

Here's another position of a different asteroid.

What do I do?

Turn right.

Let's just close that down.

We'll have to create a new job script file.

Just a plain text document called saying Neural Network Euro network.

And give that the Js extension.

Yes, and open both of those in your favorite editor arm using visual CEO code here.

The first thing we should do is set up a flag to let us know that automation is on.

So currently we have a whole bunch of flags in here.

How about we pull them out and we'll put them in a different section just to make it a little bit Tidier?

How about developer flags?

Paste them in there?

Sorry, just doing some housekeeping.

So let's create a new constant called automation automation on.

We'll set that to true initially, So when the automation is on, will want to set up a neural network So just after we set up the game parameters here, So after new game here, sit up the neural network.

For if automation is on, uh, let's leave ourselves to do here because we can't do it at the moment to dio Neural network.

Good.

Now, also, when the automation is on, we don't want the player to be able Thio Intro ll the ship, Dewey.

So down in our key press functions.

So kit key down yet Here we go, kid down and keep up.

So if the ship is dead or automation is on return, don't do anything so they can press the keys as much as I like, nothing will happen.

Similarly, in the key up method, If ship is dead or automation is on return, we'll let the computer control the ship.

Speaking of computers controlling ships, let's head down to the update function.

That's where all the objects drawn and moved, and so on.

So just the start of the update function to stop me here somewhere, how about we say, use the neural network thio control to rotate the ship and shoot and shoot?

So again, we'll just leave ourselves.

So if automation is on, we'll leave ourselves to do.

He has to dio control ship.

Okay, good.

Let's just make sure that we haven't broken anything, so I head up the top.

First of all, let's try it with the automation flag on.

Okay, so I'm moving the arrow keys.

I'm pushing the space bar.

Nothing's happened.

And just to make sure that we haven't broken anything that senator false.

Yeah, I can control the ship in that situation.

So when it's turned on, when it's true, I've lost control of the ship or soon head over to our new file, a neural network dot Js At the top of that file, let's use a flag used strict and what you strict does.

It basically enforces good coding practice things like you must declare variables and so on.

We're also gonna focus on E s six JavaScript.

Now, years six is like the latest major revision to jab scripts in tax things like class and constructors and so on.

So we're gonna use that will try to use that throughout this tutorial.

Siri's now neural networks use a lot of math, specifically matrix maths.

So how about we create a heading here called matrix functions.

A matrix is essentially just a two dimensional array.

So it has a number of rows and a number of columns.

So I'm just making to stand out a little bit.

My tricks functions.

There we go.

So happened, we represent a matrix.

Well, how about we represented as a class the class matrix?

Give it a constructor.

The constructor will take a number of rows and a number of columns.

And how about optionally Will have darter here as well.

So if we put an equal sign here, it said default value.

So the by default, it will be an empty array.

So inside the constructor, let's set the variables now, usually private variables.

Private instance.

Variables are prefixed with an underscore.

So we'll follow that convention.

Do the same sort of thing for the columns and the data, calls and data and access those variables.

We should put some getters, so just get, uh, Rose.

And all that does is return this start, underscore Rose and do a similar thing for each of the other variables.

Cole's data, this calls and this data.

So they use that when we declare an instance of this matrix.

Would go and make Chick Stock rose and that would return this stock rose.

So inside the constructor, let's initialize uh, you socialize with zeros initialized with zeros if no data provided.

So if data equals now, somebody could pass now, couldn't they or data dot length equal zero.

So, in the default case, that's true.

If either of those things occur, then we want to populate this stock data.

This Datta will equal a new empty array and let the loop through each of the rose.

So four let I equals zero.

So let's is one of these new ear six words.

Best way.

It means that Isa only visible within the four loop that Michael's here eyes less than this rise I plus plus lower case I plus plus but want to initialize wth e road.

So this starter I will equal and empty all right, and we'll loop through each off the columns That J equals zero Jay is less than this, uh, calls J plus.

Plus this data will this starter I j So we'll set each of the elements Thio equal zero Good.

If data is past will probably check its integrity.

For example, somebody could put three rows three columns but provide data with any two rows.

So that's not good, is it?

So that's what else here.

Elf school, Chick Datta integrity.

So if the daughter's length doesn't equal, the row number doesn't equal rows or dot ah, maybe dot a zero dot length.

So that's the column.

Length isn't it doesn't equal calls.

Then let's throw a new era.

So throat New Era and the era takes a message.

You could just put something like incorrect data dimensions.

That should be enough.

So let's test this.

Let's go back to our main file, Okay?

We'll have to import that for I also do that up in the style stripped credit script tags the script and we'll need a type type Feli who'll, UH text Java script and the source will equal dot sorry dot slash.

That means the current directory on neural network.

Whatever we named your file, a neural network dot Js closed the script tags, so head down to where we will initialize set up on neural network just in here.

Let's try to set up some matrices, so let em zero equal.

New Matrix rose say to rose three columns and lets out.

Put that to make sure that it's working.

We could probably use the table command here.

Is that outputs table data?

Tabula data.

So M zero dot data That's the daughter of our Matrix.

Let's take a look at that.

Just need to open the console.

You know, we have a two by three matrix full of zeros.

That's good.

Let's try toe.

Pass some data so we can do that by setting up in a ray here.

So an array of a raise I should say so to one.

Just having some values here and the next road for three 430 to say, Okay, let's give that a guy there.

We go to one minus 1430 to one minus 14 threes here.

Now What if we declared the incorrect dimension here, too?

So he said that a three by three matrix instead of two by three let's check out area checking Good era, incorrect data dimensions.

Put that back, too, also head to our other file to create another function.

So let's create a function that randomize is the whites or creates a random set of white sort of between our neurons.

There's gonna be some whites between minus one and one, so let's create a function for that.

So apply Brandon whites between minus one and one, so we'll call this random rights.

So this is inside our class four.

Let's I equals zero.

I is less than this dot Rose I plus plus for columns for Let J Equals Zero.

Jay is less than this.

Stock columns Jake Plus Plus this start Datta I J will equal math random.

That's another between zero and one times.

Two times two minus one.

So a random number between zero and two minus one is minus one between modest one and positive one.

Good.

Let's test it in our other file.

So our name zero we can discuss m zero dot random White and hopefully that should work.

So currently that's what it looks like.

Let's run it.

There we go yet that we're screwed if we got some random waits, negatives, positives and everything in between.

Great head Thio.

Other file on this created on next method Add two matrices ese so we'll make this a static method That just means we don't have to.

Instead, she ate the class we could just call it directly.

We can go in matrix dot ad.

It will require two metre sees m zero in m one.

Um, probably want to check them to make sure that they have the same dimension.

So we can go.

We'll create a new method.

A new static method matrix Start check time mentions.

It will require m zero Name one.

Let's create that now, actually, so check matrices.

These have the same dimension.

Same dimensions.

So static chick dimensions in zero and him one if ems it.

If m zero dot rose doesn't equal in one dot rose.

And what in zero dot culls doesn't equal em one dot calls.

Then let's just throw a new era, throw a new era, The message we could.

How about maitresse?

Eez?

Aah!

Off different dimensions.

Okay, so we check that now we'll need to create you.

Now admit that we'll need to create a new return matrix.

So let em equal New matrix.

The roads will be just be the same as what's past m zero dot rose m zero dot columns and really to look through each of those air.

Let's I equals zero eyes less than m dot rose I plus plus andan of lip full.

Let's j equals zero j is less than in dot culls j plus plus what we do here, so m dot data I jay will equal or this copy that m zero m zero dot data i j plus m one dot data i j.

So it is adding the each individual cells together and assigning it to the New Matrix and will have to return that matrix return in.

Okay, let's test that you now other file.

First, we need a couple more matrices to test with.

So let's create two more M one, which will be two by three.

And just change up some of these numbers here.