Bug in nightlies

[fizzix]

[QUOTE=PowerDiver;41450]One simple change would be to modify the room placement algorithm so that if a more interesting room is attempted to be placed, and the space is occupied by a less interesting room, then remove the less and place the more. I don't remember the conditions for the room placement phase to terminate, so there might need to be another change so that it doesn't just iterate until all rooms are vaults.[Quote]

Why not just start with the placement of interesting rooms. The very first calculation should be 'is there a GV on this level?' I think this is better than creating rooms and then removing them to place GVs

As to the dearth of floor items, that is yet another annoying consequence of the attack on TMJ. It used to be that you'd get their replacement from random kills of weak creatures, but with the recent change in drop level the combined changes mean that you have exchanged DL90 items for DL50 or DL60 items. It think it is much harder now to find a stat ring +6 than it used to be in 3.0, but I might just be imagining that.

I actually think the drop rates of non-consumables are fine. And not only that, with ewert's changes, the drop rates of potions are also fine. However, the drop rates of wands and rods are not. For whatever reason, I think staves are fine.

To get to dlevel 50 repeatedly without finding at least one rod of light, dstairs, dobj, dtrap and illumination repeatedly should not often happen.

[Derakon]

Starting with placement of special rooms and then placing the rest of the dungeon around them sounds like a straightforward change, but it may be quite complicated. I don't know; I haven't investigated the dungeon generation code, but such things tend to be moderately arcane.

It would probably be more straightforward to just plop vaults down on top of everything else and let the walls fall where they may, but that could create some odd-looking dungeons.

[d_m]

Starting with placement of special rooms and then placing the rest of the dungeon around them sounds like a straightforward change, but it may be quite complicated. I don't know; I haven't investigated the dungeon generation code, but such things tend to be moderately arcane.

It would probably be more straightforward to just plop vaults down on top of everything else and let the walls fall where they may, but that could create some odd-looking dungeons.

I have been working on this for awhile with no result that I can usefully show off. I would recommend taking a look at Unangband for a very different approach to the one Angband takes.

[buzzkill]

I'd agree that the autoscummer is not the answer, though I don't know what he answer is. The problem may be, and I'm just throwing this out there not having a lot of late game experience, that though DL60 and DL90 may look similar, an interesting DL60 would not make an interesting DL90.

In my early and mid game experience, I don't find the autoscummer necessary. What is necessary is to ignore the level feelings and explore the level. If you need stuff, look for it. If you're all set, then move on.

You've heard this line from me before in any number of variations tweaked to apply to any number of problems, but once again... if every level is 'interesting' then none of the levels are 'interesting'. It's a relative term. The boring levels are what makes the interesting levels interesting. In a randomly generated dungeon, I don't know what the fix is other than to make each subsequent level better (stuff) and badder (enemies) than the previous, in which case the reapers, wyrms and greater vaults will start appearing at DL20.

[Derakon]

You've heard this line from me before in any number of variations tweaked to apply to any number of problems, but once again... if every level is 'interesting' then none of the levels are 'interesting'. It's a relative term.

You're conflating "interesting" and "unusual", or at least, that's how I'm reading it. An alternate definition of an interesting level is that it's one where the player is challenged in a novel way. Nobody wants to be playing the game the same way on level 40 as on level 98. Usually this is handled by scaling the monsters you face and the frequency of unusual rooms (the appearance of graveyards creates a novel challenge in the late-game by marking areas of the map as "don't go here", for example), but it can break down.

[fizzix]

You're conflating "interesting" and "unusual", or at least, that's how I'm reading it. An alternate definition of an interesting level is that it's one where the player is challenged in a novel way. Nobody wants to be playing the game the same way on level 40 as on level 98. Usually this is handled by scaling the monsters you face and the frequency of unusual rooms (the appearance of graveyards creates a novel challenge in the late-game by marking areas of the map as "don't go here", for example), but it can break down.

except the game conveniently gives you _destruction which eliminate the graveyard threat, and pretty much any threat. Eliminating _destruction would probably make more levels interesting.

The problem is not that the challenges aren't there, but that there are so many ways to avoid dealing with the challenges.

[fizzix]

I have been working on this for awhile with no result that I can usefully show off. I would recommend taking a look at Unangband for a very different approach to the one Angband takes.

I think I have a hackish way to do this. I may try and code it this weekend, but my coding skills suck, so we'll see if I manage.

The idea would be to try and create a greater vault ONLY on the first room. Then the generation scales adjusted so that they appear at a reasonable level.

I don't think there are problems with LV, pits/nests, or unusual rooms so I'll leave these as is.

The other thing I'd like is another designation in vaults.txt that can be used for non-vaults. Just odd or interesting rooms. I'll look into adding that as well.

What do people think a reasonable GV generation chance is. I was thinking of some log scale:

Lvl 1: 1 in 5000
Lvl 33: 1 in 500
Lvl 66: 1 in 50
Lvl 100: 1 in 5

1 in 5000 is probably too high, it might be better to set an arbitrary cutoff and say no vaults below Lvl 30.

[buzzkill]

What if we (% chance but quite often) autoscummed for a really good level (contradicting my earlier post) the first time you visit a level only. Manual scumming would yield, by comparison, inferior results. After your first visit, then you're at the mercy of the boring RNG.

@ fizzix: The problem with generating vaults in a 1 in xx chance is that the % chance is rendered irrelevant by scumming. Eliminate scumming and then those numbers suddenly mean something.

[Derakon]

So you simultaneously suggest scumming and then note that another person's proposed approach only works if scumming isn't available?

[fizzix]

What if we (% chance but quite often) autoscummed for a really good level (contradicting my earlier post) the first time you visit a level only. Manual scumming would yield, by comparison, inferior results. After your first visit, then you're at the mercy of the boring RNG.

@ fizzix: The problem with generating vaults in a 1 in xx chance is that the % chance is rendered irrelevant by scumming. Eliminate scumming and then those numbers suddenly mean something.

Scumming is boring. That's the motivation not to do it. With a randomly generated game, with no permanent levels, that will always be the sole motivation not to do it. Yes, you can get to dlevel 90 and scum until you hit a greater vault with my approach. You can do it now. You can scum until BoS +10 appear in the BM. You can even scum the town for gold drops. The last two are incredibly boring though.

My approach has nothing to do with this. It merely is because the frequency at which GVs appear is lower than the code claims it should be because of the difficulty of placing them. At first attempt I'm going to try and get the GV frequency per level assuming that a GV will always be placed when it is attempted, and then go from there.


I don't care about this.

[PowerDiver]

To get to dlevel 50 repeatedly without finding at least one rod of light, dstairs, dobj, dtrap and illumination repeatedly should not often happen.

I'm not sure what you mean by repeatedly and/or not often. What's your opinion, and is there any consensus, on the probability that you find at least one of each? Are you counting mostly floor objects, or assuming you are level clearing killing 95%+ of monsters? 50% or 95% or 99% or what?

[fizzix]

I'm not sure what you mean by repeatedly and/or not often. What's your opinion, and is there any consensus, on the probability that you find at least one of each? Are you counting mostly floor objects, or assuming you are level clearing killing 95%+ of monsters? 50% or 95% or 99% or what?

I would say something like 25% of floor objects and monsters (with drops) killed. Might be a little higher since I take out pits whenever I find them.

I tend to get a higher % of floor objects once I find -dobj. Until then I am at the mercy at whether or not the alchemist has a large stack of ?dobj.

Maybe I'm misremembering, but I seem to recall finding a lot of -light, both rods and wands lying around, and they seem rarer of late. I'm certainly finding enough rings, potions, staves and scrolls. It just seems like rods and wands are underrepresented currently.

[PowerDiver]

50% or 95% or 99% or what?

I really need to be clearer. I wanted to know what minimum likelihood you think is warranted for finding all five rods you mentioned by DL50. But then I confused things by worrying about playstyles and mixing everything together.

Here's a useless datapoint. I think I found my first -dTraps in the 80s in my current game. That's diving hard enough I'd guess to miss half the floor objects, but with over 2500 kills.

[fizzix]

I really need to be clearer. I wanted to know what minimum likelihood you think is warranted for finding all five rods you mentioned by DL50. But then I confused things by worrying about playstyles and mixing everything together.

Here's a useless datapoint. I think I found my first -dTraps in the 80s in my current game. That's diving hard enough I'd guess to miss half the floor objects, but with over 2500 kills.

If I could figure out enough C coding to make a monte carlo sim, like Magnate wants, I would do it in a second. Figuring out how to get the sim to interface and log results seems way too complicated for me. Sorting through the data, though, that's what I'm good at.

[PowerDiver]

If I could figure out enough C coding to make a monte carlo sim, like Magnate wants, I would do it in a second.

I wasn't asking for the current stats. The question is the desired behavior. My guess is that what you want is incompatible with the attack on TMJ. What's the minimum probability desired to see at least one each of those five rods by DL50? If it is 50% for example, you shouldn't be so surprised to have it fail 4 [or even 6] games in a row.