Tile 0x0 in a procedurally generated world always the same

[veethree]

Hi, So I've been working on a game that uses procedural generation to procedurally generate procedural worlds. For a while, The game had some code that would find a tile for the player to spawn on by looping through whatever tiles had already spawned until it found a suitable one. But that code was written stupidly so i decided to rewrite it. In that process i set it up to always spawn the player at tile 0x0, And proceeded to try a bunch of different world seeds to find one where 0x0 is not suitable for the player to spawn. To my surprise, After trying about a 100 different seeds i wasn't able to find one. The 0x0 tile always seems to be a ground tile.

Is there any reason this should be the case? I looked over my world generation code and couldn't come up with any reason why that should be the case.

The game decides where ground tiles go via this function

Code: Select all

function generateNoise(x, y, scaleBase, scaleDetail, thresh, ratio1, ratio2, seedOffset)
    scaleBase = scaleBase * noiseScale
    scaleDetail = scaleDetail * noiseScale
    return noise(x * scaleBase, y * scaleBase, seed + seedOffset) * ratio1 + noise(x * scaleDetail, y * scaleDetail, seed + seedOffset) * ratio2 > thresh and true or false
end

where x and y are the tiles grid coordinates, scaleBase and scaleDetail scale the two layers of noise, thresh is the threshold for it returning true, ratio1 and ratio2 are the ratios of the two noise layers and seedOffset is a value added to the base seed, Which is used as the third argument in love.math.noise.

The full world gen code is here(github). Along with all the other code.
The code on github still does the whole finding suitable tile dance. The .love below does not.

[milon]

You're using Love's simplex noise function, which has the following notable constraints:
1. Given the same inputs it will ALWAYS produce the same output (it's a stateless generator)
2. Return values tend to be constant given integer only inputs

What do the inputs evaluate to for cell 0,0? I'm guessing a seed-independent constant, or else an integer that results in a constant output. (I don't have time just now to dig into that, but you should be able to work it out quickly.)

[veethree]

The first 2 arguments for 0x0 will always be 0, but the seed, which is used as the third argument, will be different in each world. And the seed is always an integer. I don't have time right this moment to test it, but this makes some sense. Thank you!

[milon]

Yeah, makes sense then. Coords 0,0,[INT] seems to always give 0.5 as an output, which your map generator must interpret as a ground tile.

Here's a quick example:

Code: Select all

for i = 1, 5000 do
    print(love.math.noise(0,0,i)) -- always gives 0.5
end

Changing the 0 to 0.5 results in "better" noise values, but you still get 0.5 for many of the values. Try to use non-integer values for better noise values. You can divide x and y by 10, or add pi to them, or something else that will (usually) give a non-integer result.

But hey, if you're happy with what you have, even better!

[veethree]

Yeah, makes sense then. Coords 0,0,[INT] seems to always give 0.5 as an output, which your map generator must interpret as a ground tile.

Here's a quick example:

Code: Select all

for i = 1, 5000 do
    print(love.math.noise(0,0,i)) -- always gives 0.5
end

Changing the 0 to 0.5 results in "better" noise values, but you still get 0.5 for many of the values. Try to use non-integer values for better noise values. You can divide x and y by 10, or add pi to them, or something else that will (usually) give a non-integer result.

But hey, if you're happy with what you have, even better!

The x and y are usually floats, Because they're multiplied by some value between 0 and 1. I did try making the seed a non integer and it seems to result in slightly more interesting worlds. It's kinda hard to tell.
Thank you for your help!

[zorg]

Basically, integer coords are grid points, so their values will be nearly, if not completely the same; also, even if you multiply by some float, 0 times any value is still 0 you know

adding an offset of .5 should be a simple solution that mitigates this:

Code: Select all

function generateNoise(x, y, scaleBase, scaleDetail, thresh, ratio1, ratio2, seedOffset)
    scaleBase = scaleBase * noiseScale
    scaleDetail = scaleDetail * noiseScale
    return noise(.5 + x * scaleBase, .5 + y * scaleBase, .5 + seed + seedOffset) * ratio1 + noise(.5 + x * scaleDetail, .5 + y * scaleDetail,.5 +  seed + seedOffset) * ratio2 > thresh and true or false
end