Why require probabilities > 0?

[butlersrepos]

I am leveraging the custom data probability to use WFC to generate various biomes, some of which are made from the same tilesets.

I do this by adding a new custom data for each biome, like "Plains", "WoodedPlains", "RiverPlains" and setting different probabilities per biome.
Example, water tiles in "Plains" may be rare (0.05) but in "RiverPlains" they would be higher (0.2).

My issue is that there is an assertion that all probabilities must be >0 however the Godot TileSet will default every tile to 0.0 if you add a new custom data layer as floats. I prefer this since I can start from a selective setup and enable tiles per biome as I design them.

So far I am just commenting out this assertion but I wanted your thoughts on if this was a reasonable approach that I'm employing? and if so should I just continue to leave this assertion commented out? :/

Thank you! This plugin is amazing and I'm having a huge amount of fun with it. I appreciate your efforts so much.

[AlexeyBond]

Hi.

Your approach is reasonable, at least if there may be different bimes on a single map - in such case the generator should be aware of all tile types that could be generated in other biomes (including tiles that have probability of 0 in the current biome) in order to generate map near biome edges.

But if a tile map always contains only a single biome, it would be more efficient (performance-wise) to have separate rule sets (samples + generator settings) for each biome that don't contain tiles not present in this biome (they may still share single tile set - set of tiles used by the generator is determined by what's represented in the sample map; and use different data layers for probabilities). Extra tile types (even ones with probability of 0) increase complexity of WFC computation. However, there's a computer performance/human performance tradeoff, so your current workflow may still make sense if it takes too long to make and maintain different sample maps for different biomes. It's should also be possible to create some sort of automation to generate different rule sets for different biomes (subsets of tiles) based on a single sample map.

It would be a bit safer to change assert to probability >= 0.0 instead of disabling it. The tile choice function (pick_divergence_option()) behavior will be almost adequate with zero probabilities (tiles with 0 probability will not be chosen if there are tiles with higher probability, when there are only 0-probability tiles left - the first one will be chosen), but it may get broken with negative values. I think it would be better to pick a random tile when there are only tiles with 0 probability left - but that's one of small changes I thought of, but haven't implemented yet.

[butlersrepos]

For clarification: when you say

"Extra tile types (even ones with probability of 0) increase complexity of WFC computation"

do you mean only the ones we used in creating the sample map? as in, just don't actually draw them anywhere in the sample?

[AlexeyBond]

The generator will only use the tiles present in the sample map. If there are any tiles that are present in the sample map but have a probability of 0, then the generator most likely won't place them (unless there is no other choice) but it still will be aware of those and that "knowledge" will decrease performance a little.

If a tile has an invalid probability value but is not present in the sample map then the generator won't consider such tile and thus it will not fail the assertion.