PDA

View Full Version : Stats


Magnate
October 28, 2011, 10:57
I thought people might be interested in an early comparison of stats between 3.3.0 and v4.

For artifacts, current v4 is actually slightly more generous for DLs 1 to 7, then significantly less generous for the rest of the game (the y axis is "artifacts per game at this dungeon level"). It's amazing that two such different algorithms produce numbers that are even the same order of magnitude, let alone vaguely the same shape. I was expecting it to be much more different. It may still be too stingy (it's about twice as generous as it was a few days ago), but let's see how it plays. The optimum distribution is probably between the two lines.

For egos, the picture is significantly more complicated. First, we have to assume for the graph that each item gets three affixes or one theme (I haven't yet worked out how to calculate an average number of affixes per item, but I suspect 3 is not too far off). Second, we have to bear in mind that lots of affixes are nowhere near as good as the old egos - there are currently *no* bad or cursed egos in 3.3, whereas a number of affix items will be pretty much unusable because of Broken or Damaged. Third, I've added the themed items to the affix items, but they are only 1-10% of the total (<1% at shallower levels, rising smoothly to 9.7% at dl100). The stats don't tell us how many of the 3.3 egos are of the "high-end" variety now found as themes(*), but the overall impression is that v4 provides a higher quantity of ego items but with a smaller proportion being really good.

Which was kind of the point. (It's quite possible that affixes will need toning down deeper in the dungeon, but nobody's got that far yet.)

(*) Actually they do, but my SQL-fu isn't up to that yet.

I've just noticed that the colours are reversed on the two graphs. Just to keep you on your toes. Sorry.

EDIT: It's also worth pointing out that the affix items include all those non-egos which 3.3.0 would have termed merely "good", and will therefore not show up in the 3.3.0 counts. To enable a proper comparison I'd need to add "good" items from 3.3.0's stats. Off to work out how to do that ...

Magnate
October 28, 2011, 17:32
So, I managed to construct the hideous query to extract the data on the 36 "high-end" egos from the 3.3.0 stats. (I also added the figures for MHDSM, Balance DSM and PDSM, so we can compare against all the themes in v4.)

The first graph looks pretty ugly, but it's not too bad - there's a reasonable correlation in about half of the cases. The big discrepancies are:

More than 4x as common in 3.3.0:
85x *Slay* Orc (#26)
27x Magi cloak (#7)
24x PDSM (#39)
13x Lordliness (#10)
9.7x *Slay* Troll (#27)
9.6x BSDM (#38)
8.3x Blessed (#35)
6.9x Magi crown (#8)
6.1x Haradrim (#32)
5.7x Dwarven armour/shields (#5)
4.9x Might (#9)
4.9x Westernesse (#20)

More than twice as common in v4:
3x Earthquakes
3x True Sight

This threw up some very interesting factoids:

PDSM appears approximately once every ten games in 3.3.0. You may well not see it, of course, but it's generated somewhere in levels 1-100. Balance DSM appears on average every other game!

Using alloc_max is hugely effective in limiting the generation of dross. This is the reason for the lack of *Slay* Orc/Troll weapons in v4.

The distribution curve of these items throughout the dungeon is astonishingly smooth in both cases (though something odd happens in v4 past dl90). Moreover, the relationship between the two curves is amazingly consistent, given their completely different routes to these items. See the second graph.

artes
October 28, 2011, 23:16
Diagrams are cool.

fizzix
October 28, 2011, 23:35
These look cool!

I was thinking about reworking wiz-stats, and I may still at some point in the near future. However, the stats module is probably better suited at answering these questions, which I think are useful for the head to head comparison.

Damage (and to-hit) is probably the trickiest thing to even out with affixes. What do the curves look like for?

weapons with +5 to dam or greater (especially in dlevels 1-10)
weapons with +10 to dam or greater
weapons with +15 to dam or greater
weapons with +20 to dam or greater
weapons with +25 to dam or greater

endgame quality weapons:
needs +15 to dam or greater, big dice (average 18 or greater, 18 is BoC in 3.3.0) and slay evil. Obviously we have many more options with dice in v4, so this should be an interesting comparison.

If you can't figure out how to SQL this, I'll try to put it in wiz-stats.

Derakon
October 29, 2011, 01:59
I don't think requiring 18 average from dice and slay evil is a requirement for an "endgame quality" weapon. What matters is your average damage/round against Morgoth; I'd say if you can manage at least 400/round (per the 'I'nspect screen, which of course ignores misses and monster AC damage reduction) then you're in good shape.

Of course, that's harder to craft a query around...

fizzix
October 29, 2011, 03:55
I don't think requiring 18 average from dice and slay evil is a requirement for an "endgame quality" weapon. What matters is your average damage/round against Morgoth; I'd say if you can manage at least 400/round (per the 'I'nspect screen, which of course ignores misses and monster AC damage reduction) then you're in good shape.

Of course, that's harder to craft a query around...

18 average from dice is a blade of chaos. SoS and MoD are both more. But the ease of the query is the main reason. It's still a bit unsatisfactory, because it may be possible for v4 to create a 50 pound weapon that you only get 3 blows max from.

Derakon
October 29, 2011, 07:57
Right, what I'm saying is that requiring 18 average from dice will exclude a lot of endgame-quality weapons. I'd say more like SUM(to-dam + (dice + dice * sides) / 2) > 50. That might be a bit high; I don't have a good estimate for how much damage comes from off-weapon sources at the moment.

Timo Pietilš
October 29, 2011, 11:15
18 average from dice is a blade of chaos. SoS and MoD are both more. But the ease of the query is the main reason. It's still a bit unsatisfactory, because it may be possible for v4 to create a 50 pound weapon that you only get 3 blows max from.

18 is Aule. Eonwe, Durin, Ringil, Zarcuthra etc. have smaller dice. I think only Deathwreaker, Doomcaller, Pain and Aule reach that 18.

Try 4d4: 4*2.5 = 10. 18 is way too much.

fizzix
October 29, 2011, 15:11
18 is Aule. Eonwe, Durin, Ringil, Zarcuthra etc. have smaller dice. I think only Deathwreaker, Doomcaller, Pain and Aule reach that 18.

Try 4d4: 4*2.5 = 10. 18 is way too much.

I specifically do not care about artifacts though. If we were testing the randart code, this would be important to consider. I maintain that BoC, SoS and MoD all have average of 18.

Magnate
October 29, 2011, 22:17
I specifically do not care about artifacts though. If we were testing the randart code, this would be important to consider. I maintain that BoC, SoS and MoD all have average of 18.Fortunately we do log the +dam of every item by kind, depth and origin, so this query will be doable (albeit syntactically difficult for a novice). I'll see if I can do it during my next dull day at work (Monday).

Antoine
October 29, 2011, 22:23
Fortunately we do log the +dam of every item by kind, depth and origin, so this query will be doable (albeit syntactically difficult for a novice). I'll see if I can do it during my next dull day at work (Monday).

Will your next dull day at work be Tuesday by any chance?

A.

Magnate
October 29, 2011, 22:28
Will your next dull day at work be Tuesday by any chance?Probably - though they threw me a curveball on Friday and asked me to draft a Board paper in three hours. That was pretty interesting.

Magnate
October 31, 2011, 14:29
So, I ran the consumables stats, and the headline is that v4 is slightly more generous across the board, by virtue of there now being slightly fewer weapon and armour base items to compete with them. There is a baseline differential of about 15% which is the case for:

all stat potions
rods of Detection
wands of Annihilation
wands/rods of TO
mushrooms of Vigor
rods of Healing
staves of Banishment
staves of the Magi

But there's a significant increase in all the things which have been given the OF_GOOD flag:

rods of Speed
potions *Enlightenment*
scrolls of Rune of Protection
potions of Experience
scrolls of *Acquirement*
potions of Life
scrolls of Mass Banishment
potions of Augmentation
scrolls of Banishment
scrolls of *Destruction*
potions of *Healing*
scrolls of Acquirement

... this varies from about 2x (the bottom half of that list) to 5x (rods of Speed are stupendously rare in V). It's clear that the OF_GOOD mechanism promotes these things to be much more findable - I hadn't realised how big a proportion of drops must be "good" or "great". Alloc probs of these items will need to be adjusted at least in line with V (and I think there's a consensus that 3.3.0 is a little too generous, though we don't want to go back to 3.1.x levels).

Bizarrely this trend doesn't hold for the dungeon books, which are about 25% *less* common in v4, despite getting the OF_GOOD flag. I'm not sure why that is, but it's easy enough to correct if they are now too rare.

N.B. The mechanism for choosing a base object kind hasn't changed - the only changes are that some base objects are no longer findable because they're now provided by affixes or themes (DSMs, MoD, SoS etc.), and some consumables got given the OF_GOOD flag so they don't get excluded from "good" and "great" drops.

Magnate
October 31, 2011, 14:34
The distribution curve of these items throughout the dungeon is astonishingly smooth in both cases (though something odd happens in v4 past dl90).I found this last night: the ego_item.txt for v4 had a bunch of ill-thought-out alloc_max values which were less than 100, which meant that around DLs 85-90 a whole bunch of affixes stopped being available. This meant that the related themes suddenly got much less common, hence the dip at the end of the graph.

I'm just about to release a new version of v4 (one bug left to fix) which sorts out the knowledge menus and allows squelching of affixes and themes. This version will have these maxima corrected to 100 for the time being, until we get more reports of how the endgame plays. I'll also make some adjustments to the commonness of the "good" consumables (see other post).

buzzkill
October 31, 2011, 14:55
@ Magnate: Kudos on v4. I can actually get excited about some of your wacky :) ideas now that they are be developed in a pseudo-variant environment. The whole affix thing is looking very cool.

Derakon
October 31, 2011, 16:17
Spellbooks likely became less common because they were predominantly showing up in guaranteed-good drops, which are now more diverse due to the inclusion of those now-too-common good consumables. :)

Make one thing more common, it makes everything else less common...

Magnate
October 31, 2011, 20:20
Spellbooks likely became less common because they were predominantly showing up in guaranteed-good drops, which are now more diverse due to the inclusion of those now-too-common good consumables. :)

Make one thing more common, it makes everything else less common...Of course. Before OF_GOOD, spellbooks were about the only non-wearables available from "good" drops, so of course they'd be rarer.

Yes, balancing is difficult. I don't want to flood the dungeon with ego items when scaling back the consumables. And I don't want to get into too much detail balancing affixes when there are so many planned changes to monster distribution and combat! But at least we have something vaguely playable in the meantime.

@buzzkill: thanks ;-)

fizzix
October 31, 2011, 20:28
With regard to dungeon books, I recommend we keep the current probs for MB8 and MB9 as exist in Vanilla.

Namely:
There should be roughly a 94% chance of finding M/PB9 by clearing every level once.
There should be a near 100% chance of finding M/PB8 by clearing every level once.

This test is easy to run in wiz-stats and I can rebalance this whenever you want me to (it will constantly need readjusting as we continually adjust other parameters)

Not being able to find M/B9 as a mage or a priest is not much fun.

There will still be a problem of too many copies of books 5,6 and 7 showing up. But I don't think that's resolvable except by making the dungeon books artifacts.

Derakon
October 31, 2011, 20:47
There will still be a problem of too many copies of books 5,6 and 7 showing up. But I don't think that's resolvable except by making the dungeon books artifacts.

Which would be great if for no other reason than so I only have to throw away Holy Infusions once. :)

Magnate
October 31, 2011, 21:37
Which would be great if for no other reason than so I only have to throw away Holy Infusions once. :)It's on the list. Since it will mean yet more rebalancing (another ten base objects gone), it might as well be sooner rather than later.

Magnate
November 2, 2011, 10:51
Here are the graphs for to-hit and to-dam values. They are, as you would expect, very similar in shape. Overall they show that v4 produces significantly lower hit and dam values than V - the area between the two lines is probably about 50% of the total. Of course, V's generation is not open-ended, so values over 35 are never generated. v4 generates a tiny number of values up in the 40s and early 50s, until it hits the hard cap in object.h - but for some reason only for to-dam (it never generates to-hit over +40). This may or may not unbalance the endgame - we'll see when somebody gets there.

Two things to bear in mind about these graphs:

1. They include all wearable items, including rings of accuracy/damage/slaying/reckless attacks, gloves of slaying/power etc.

2. They don't show the data points for +0 (because that renders the graphs useless). v4 generates 29% more items with +0 to-dam and 35% more items with +0 to-hit - but those percentages are deflated by the presence of all the armour items. When I construct the query to distinguish between weapons and armour, I suspect we'll find that v4 produces nearly twice as many +0 weapons.

fizzix
November 2, 2011, 14:48
Cool thanks. What are your opinions on these. Do you think v4 is too stingy, or just about right?

Magnate
November 2, 2011, 15:34
Cool thanks. What are your opinions on these. Do you think v4 is too stingy, or just about right?I don't have a view yet - I fear the worst of both. People will be stuck with weapons with less than +10 for way too long, and will then suddenly find something with +40 and everyone will complain that it's horribly broken.

The first thing I thought was that we really need to make +dam give diminishing returns - but since that's the whole reason why we don't just move to O-combat, that's a non-starter. Balancing generation would be easier, but the cost in terms of intuitiveness would be too high.

I do think we should start making some combat changes before worrying too much about fine-tuning ego balance though. Things like reworked to-hit and critical hits could make quite a big difference.

It's probably also worth separating out on-weapon and off-weapon hit/dam. I'll have a word with myshkin - it might be easiest to do this at collection time with separate tables, rather than trying to set up queries to do it.

Magnate
November 2, 2011, 16:22
Fortunately all the weapons are contiguous in object.txt - index numbers 30 to 122 inclusive. So it was easy to separate them out. The graphs look roughly the same shape as before, but the conclusions are worse: it means that hit/dam on nonweapons is almost identical (as we would expect), but v4 is much meaner with weapon bonuses. In particular, v4 generates 15x as many weapons with +0 to hit, and 13x as many with +0 to-dam. (Again, I didn't show these in order not to compress the graphs.)

Now, this difference is offset quite a lot by the fact that +0 weapons in v4 can have slays, brands, extra dice and all sorts of other niceties which +0 weapons in V cannot have. But I don't think that quite makes up for the meanness. I'm not sure that this is something stats can test (I don't think we can track the hit/dam on weapons with specific flags - I'll ask myshkin), so we'll have to see how it plays.

Jungle_Boy
November 2, 2011, 21:58
What seems to happen to me in my games is that I am stuck with a crappy +2 or +3 weapon, then I get lucky and find an artifact with +10 dam which is much better than anything else coming along for quite some time. I'm not sure how long cause I always end up dead before then but I would say down to about 2000 ft at ealst (that's where I die)