# Procedural generation

Procedural generation is a technique used by programmers, typically used in games. It essentially refers to creating assets like textures and terrain algorithimically instead of manually.

It is used widely for all sorts of things, from creating simple linear graphs of curves to making heatmaps and randomized textured terrain. In this guide I will explain how to use Perlin Noise, a technique that generates smooth curves that can be used in all manner of ways.

# Perlin Noise

Perlin Noise originated in the 80's when a guy named Ken Perlin decided he was sick of the computery look graphics had. He created a basic program that allowed him to generate decent looking textures, like the one below:

There are a few differences, though, as that that version uses a later version of his noise ported to JS and it is also animated. However, that's not the point; the point is that Perlin Noise is absolutely amazing, as you will see later.

In this post we will look at 1D Perlin Noise.

# 1D Perlin Noise

To begin with, I'll explain what Perlin Noise consists of. There are two parts for a basic 1D implementation. A psuedo-random number generator, and an interpolation function. Put together with some canvas magic, it looks like this:

Cool, huh?

## PRNG (Pseudo-Random Number Generator)

Essentially a PRNG generates numbers that may seem random but actually aren't. If you give the same starting value to a PRNG it will return the same result. So, if you want to generate random numbers using a PRNG, use a random number to seed it. Please note that I made a typo in a lot of the pens (PSNG instead of PRNG) so just ignore that if you see it.

You're probably asking why we're using a PRNG and not good old `Math.random()`. It's simply because you can save a seed and get the same result. Plenty of games depend on this. Take, for example, Minecraft. When you generate a world, you can choose to input a seed for the world to be based off. And when it comes to the generation of that world, can you guess how the world is made so smooth? *Cough* Perlin Noise *Cough*. You get the idea.

Now, we will actually make a PRNG. We'll make a Linear Congruential Generator, or LCG for short. It is an especially simple PRNG, and only requires a few lines of code to see the magic in action.

```  ```var M = 4294967296,
// a - 1 should be divisible by m's prime factors
A = 1664525,
// c and m should be co-prime
C = 1;
var Z = Math.floor(Math.random() * M);
function rand(){
Z = (A * Z + C) % M;
return Z / M;
}
```
```

Don't worry about the huge numbers there, they are the parameters for our LCG. I chose large numbers because the length of the sequence of numbers you get is equal to M. So the bigger M is, the more possible values you can have.

Now, this generator returns decimals between 0 and 1, just like `Math.random()`. It can be seeded, although in that example we've seeded it with a random number. Now we're ready to move onto interpolation.

### Interpolation

Interpolation is the act of creating new data points between two given ones in a way that makes the data look smooth. I suppose that's not a particularly precise definition, but it'll do for what we're doing. We are going to interpolate between two points (generated with our PSNG) using cosine interpolation. It looks like this:

```  ```function interpolate(pa, pb, px){
var ft = px * Math.PI,
f = (1 - Math.cos(ft)) * 0.5;
return pa * (1 - f) + pb * f;
}
```
```

You can try and work out how that works if you want, but there really isn't much point. In essence, this function takes three parameters, a, b and x. A is the first point, b the second, and x, a decimal between 0 and 1 specifying where along the space between the points the point who's value you want. Now, that's a rather complicated example, though the results make the line look smooth, as in the last example.

```  ```function interpolate(pa, pb, px){
return  a * (1 - x) + b * x;
}
```
```

That interpolation function is linear, so your lines will look jagged. Unless you're using Perlin Noise every frame in a CPU intensive game, you shouldn't use a linear function. Even then, try not to if possible. The example is only there so that you can see how it works without the trig bulking it up.

## Putting it all together

We'll define a few variables first, which will be the starting point for our noise. Feel free to change them around.

```  ```var x = 0,
y = h / 2,
amp = 100, //amplitude
wl = 100, //wavelength
fq = 1 / wl, //frequency
a = rand(),
b = rand();
```
```

The amplitude, in case you don't know, is the distance from the top (or highest possible value) to the bottom (or lowest possible value) of the wave. The wavelength is the distance from the peak of one wave to the peak of the next (that is, along the x axis, if the wave is oriented the same way as in the previous example.) The frequency, although we don't use it in the end, is the amount of waves in one second (or other unit of time).

Now that we've got the setup, let's go crazy!

```  ```while(x < w){
if(x % wl === 0){
a = b;
b = rand();
y = h / 2 + a * amp;
}else{
y = h / 2 + interpolate(a, b, (x % wl) / wl) * amp;
}
ctx.fillRect(x, y, 1, 1);
x += 1;
}
```
```

That plots the line along the x axis on a canvas, using the variable `ctx` as it's context. Now, we put that all together with some canvas setup and we have the example I showed you earlier! Now that we have basic 1D Perlin Noise plotted, we should move onto having multiple octaves, creating the effect below:

## Octaves

As you can see in that example, you can make lines that are very bumpy. The way we do this is by adding new octaves. Essentially what this means is that you get some Perlin Noise. Then, you divide the amplitude and wavelength by a divisor of your choice (2 is probably the best option), and then we make another lot of Perlin Noise. That second line is another octave. You should probably make enough octaves in your lines as can be seen on your canvas (so until your Perlin Noise values are less than one). With an amplitude and wavelength of 128 we can have about 8 octaves.

So, once we have our 8 octaves, each a progressively smaller curve, we add the values of each together. Don't average them, as that means each additional octave we add brings the resultant curve closer to 0. So, we add them together and get the values for our awesome bumpy curve. Here's my implementation of this code, coming straight out of the example above: (the `Perlin` function is a 1D Perlin Noise making function, using all the stuff we've been working on. It returns an object, where the `pos` property is an array containing all the values.)

```  ```//octave generator
function GenerateNoise(amp, wl, octaves, divisor, width){
var result = [];
for(var i = 0; i < octaves; i++){
result.push(new Perlin(amp, wl, width));
amp /= divisor;
wl /= divisor;
}
return result;
}

//combines octaves together
function CombineNoise(pl){
var result = {pos: []};
for(var i = 0, total = 0, j = 0; i < pl[0].pos.length; i++){
total = 0;
for(j = 0; j < pl.length; j++){
total += pl[j].pos[i];
}
result.pos.push(total);
}
return result;
}
```
```

Add a line drawing function and you're ready to go! The final code would look like this, assuming that we've written a `DrawLine` function:

```  ```DrawLine(CombineNoise(GenerateNoise(128, 128, 8, 2, window.innerWidth)));
```
```