Idea - autogenerate branches!

[danr]

I was thinking, I bet an algorithm could be developed that would generate branches, not just dungeon levels. So instead of having fairly set branches, the game could generate entirely different branches each time.

It would start with just one branch to see how it goes - the game would pick a monster set, e.g. trolls and dogs or something, generate 4+1d3 levels of a random style (open, closed, watery, etc.) and the bottom would have loot that is balanced to the toughness of the branch.

I know, another outrageous idea from danr...

[seth]

I think balance would be the problem here, especially if many branches were randomized. I do like randomization, though.

[zasvid]

I imagine it would be hard to autogenerate good flavour of branches. Consider randarts -> their names usually have little to do with what they represent. Now extend that problem to characterisation of branches. Results probably would be quite silly, like City of the Hydras or the Dwarf and Elf Beach. Sacrificing flavour shouldn't be an option, because it is quite important, at least for alluring new players with fantastic locations. I know that when I visited the Lair or the Swamp for the first time I was amazed by the wonderful, colourful and distinguished look and feel of those places.

[danr]

I would love to find a "silly" branch. There could be a dedicated thread for it here - "What kind of crazy autobranch did you find?"

There would certainly be challenges in doing it well.

I just feel like the current set of branches is too predictable, even with the random alternation of lair sub branches.

Also - I thought there was supposed to be a dwarven halls branch? I haven't seen it yet.

[zasvid]

I would love to find a "silly" branch. There could be a dedicated thread for it here - "What kind of crazy autobranch did you find?"

Well, I would hate to - I am a very serious roguelike gamer

There would certainly be challenges in doing it well.

I just feel like the current set of branches is too predictable, even with the random alternation of lair sub branches.

Well, I think that it would be better to have more branches randomly alternating themselves between games and continue trying to make existing branches more interesting before tackling a mountainous effort of a random branches.

However, the idea is not without merit and perhaps addition of one random branch could provide the testing grounds for this. Maybe just change one of the existing branches to provide randomly themed levels first, e.g. change Pandemonium into a Realm of Chaos that's different every time you see it.

Also - I thought there was supposed to be a dwarven halls branch? I haven't seen it yet.

They seem to still be working on it.

[danr]

I agree it could take work, but not necessarily any more than just designing another branch from scratch. And once the algorithm is refined the up-front work would have essentially unbounded payoff.

I also agree it would be best to start with one branch being like this just to test the potential.

Basically I see it as an extension of the roguelike principle of infinite replayability being generated from a core set of algorithms. It's the randomness of the dungeon that creates the replayability, this just extends that.

It would be pretty easy for an algorithm to do the balancing, perhaps even easier than for a human designer. You can easily estimate the total XP value of the monsters likely to be generated, the challenge level by the average HD of the monsters etc., and loot can be easily matched to be appropriate to the challenge level.

In fact this could be built in as a late-game balancing tool - if the player has had too easy of a ride until then, it could be weighted to being really tough. If the player has randarts in every slot and six good mutations, the level could be a single-level branch filled with floating eyes, or the same as normal but with burning walls (like acidic but with sticky flame effect). If the player has had a really hard slog and has only found three rings all game, it could be weighted to have a bit more loot.

If I WERE a dev and had the time for this, I would see this as a really fun challenge

[dpeg]

danr: It takes a lot of work to set up a branch. I think we needed three release cycles to get the Shoals done. A new branch is more work than a new god or a new species. Also, a branch algorithm would produce inferior branches. (This has nothing to do with branches per se: handmade vaults are better than randomly produced ones; I love the idea of random gods, but they will obviously be less consistent than the designed ones etc.) Of course it's obvious that the whole genre is driven by randomness, we are very aware of that. But that does not mean that randomisation is useful for everything. (To give a minor example: starting stats used be randomised, now they're fixed.)
Sorry to spray water, but unless someone really, really likes this idea, it will never materialise. And I don't mean a developer here. Some pointers what should be considered: A branch should have a certain monster set, threat (this is related to, but more than monsters), theme, loot (this includes total xp), layout. On top of this, it should be distinguished from branches already in place.

[MrMisterMonkey]

written before dpeg:

I think the biggest challenge, aside from ensuring sensible branches, is generating a good branch ending, unless this sort of thing doesn't end up as a real branch, in which case it's no worry.


written after dpeg:

I agree with dpeg.

[danr]

I think it would be a really interesting challenge just to see how well it could work. I'm more interested in that question than I am in the gameplay effect of a randomized branch.

But I do realize that unless someone who actually knows Crawl's source code also thinks this would be an interesting experiment, there's no point pushing for it.

I know it's a challenge. That's what interests me in the idea. Well, thanks for humouring me anyway.

[KoboldLord]

Most of the current branches only work well because their inhabitants synergize with the terrain. Imagine randomly generating Shoals, only with non-aquatic inhabitants. Compare fighting Saint Roka & Friends in Orc 4 vs. fighting Saint Roka & Friends in Shoals 2, and you'll see how important this is. Dropping a damaging cloud on him and giggling while levitating far away over the water where he can't get at you is fun when it happens once and you remember all those times you fought him in Orc 4 and this is payback, but would you want to clear an entire branch this way?

[danr]

Well, that would be part of the challenge, but not a major part. It could easily be programmed to only choose an aquatic monster set iff the branch has rolled an aquatic terrain form.

Perhaps "random" was not the best word. Random but appropriate - that's the challenge. Creating a non-problematic theme would not be that hard. Balancing it with the rest of the dungeon, that's the hard part.

Although the design challenge is what most interests me, I do see a certain gameplay benefit. Currently one pretty much knows that you want certain resistances and abilities for certain branches, and players gear their characters toward that. With an "autobranch" players may get something that they will have to play very creatively to make their character work in it.

[dpeg]

I don't think that branches are the best playground for this. If you want monster sets, look into ziggurats. If you want randomised but thematic levels, this could fit Pan.