My idea will have to wait a while but my first thought on this thread is that I'd love to take the json planets + the user ratings and use them to train a ML model so that I can make an engine that automatically identifies/generates "good" planets. ...or maybe Sorian is already on that?
It's really easy. You get the community to vote on whether a planet is good or not. If you assume that good planets are played more often than bad planets, you could just train off the frequency.
Yep. The hard part is getting enough data (ratings and/or play frequency), that's why this idea would probably have to wait until the game has been released to enough people. If this were built into the game (or added as an update), the engine could generate a few random json outcomes of a seed and then only build the "best" one for the user.
What you're doing with the mod-ability of PA is really awesome. Are you looking to have any of the interface modding possible during alpha? Even if there's minimal to no documentation, it'd be pretty cool to be able to play around. Edit: Actually, another question. Will any JS libraries and frameworks (such as Prototype JS and/or jQuery) be included out of the box?
What sort of size are these Json files? For example the one for the planet shown in the 3rd may livestream? (or one like it)
One problem with anything learning is that you need to be able to control output of your planet. Not just that it's repeatable, but in a way I don't have a good term for. So lets go with an example: Can you given a seed (and identical generator engines) and a output planet, and you want something with slightly less mountains, but them all in roughly the same spots with most of the planet the same? Can you tweak the seed in a known manner or does this require randomly picking seeds till it finally pops out? So i guess it's: Can you map the output(planet) to the input(seed)? This all I guess depends on what exactly the seed and the generator engines look like. I am quite intrigued though.