LK vs Travincal vs Cows - runefinding guide

[pharphis]

One of the problems with locked patterns is that if the HR is behind the locked pattern and not the unlocked one, then not only is it much more comparable to normal drop rates (I don't actually know, but my guess is LK is 10x better than monsters for some HRs, and definitely so in older versions when a pattern exists) but it also is extremely difficult to root out those patterns. What it requires is very careful monitoring of the sub-pattern excluding the HR... which is unlikely because no one looks closely at patterns unless there is an HR

Alternatively, we could collectively comb through item find thread/LK thread for pattern confirmation rate. Those that come up unusually little may require further scrutiny and focus. Probably at least one HR falls into this category

 

[Brak]

@Brak very interesting theory. This would mean, in essence we have 128k patterns.
65k locked, 65k unlocked

no it doesn't work that way
The locked chest patterns get tacked onto the existing 65k pattern. They just add one fixed drop cycle to the already documented pattern. So the number of locked 'extended' patterns would roughly equal the number of patterns from normal chests (which are funnily enough just truncated versions of the superchest patterns).

tldr; its turtles all the way down

 

[GalaXyHaXz]

That would make sense, but it still means we have 65k drop patterns with 65k additional items per locked chest. Some runes might pop up upon the locked chest only, which is why some might be a PITA to find. I don't feel like testing anymore, but Its good to keep an eye out in the future. I would guess a locked chest is going to have more runes.

 

[Grape1]

which are funnily enough just truncated versions of the superchest patterns

is this only chests, or other objects too btw?

I vaguely remembered this being discussed, maybe in the RoF thread last year, and it got me wondering as I found couple of Act5 Zod drops:

both should be P8.

Does that indicate that A5 superchests have a Zod pattern? Now someone remind me, where were those located?

 

[pharphis]

A5 superchests aren't "sueprchests" in that they would have 65k patterns, to the best of my knowledge. They aren't the same object type as earlier acts.

They more likely have 4 billion patterns like most other objects... in which case, of course they have many zod patterns.

 

[GalaXyHaXz]

What pharphis said is right, unfortunately act 5 has no 16 bit chests. The chest object -580 spawns in act 1-4, and always has the "magic" flag set. Chances are if it did spawn in act 5, there might be one (1) Zod pattern. Your finds were sheer luck, and congrats. I've seen every rune drop on my toons except Cham/Zod >

Edit: if you want to talk about winning the lottery, this guy takes the cake. Note this was pre-1.13 when runes were 10-15x more rare.

 

[darkstarhub]

Great guide and awesome information guys. I love statistics, but those are some long odds when you aren't the fastest of runners. The one thing that saves me is luck I guess. When I read about somebody running a series of a couple thousand LK runs and not finding more than a Mal I think to myself, there is no way I would keep doing runs the same way if that's all I found. I realize statistically over time odds will usually even out, but extremely bad stretches of luck are tough to deal with.

I compare hitting LK chests only to running Pindle. It doesn't take me long to feel like my pinky finger is getting sawed off with a butter knife. That's why I like to throw in a little side action on the runs, whether it be hitting certain poppables or killing a monster pack. If my luck is poop just hitting the chests I usually do a few runs of chests and just about every poppable along with most of the monster packs and most of the time I get something. The usual sets go about 30 runs chests, 30 runs hitting some extra stuff, repeat a few times, and if nothing decent drops do a few longer clears.

Will I do better over time statiscally by changing things around? Probably not, and I know it's not as efficient, but my skill as a D2 player does not come close to my love for the game. Some of you guys are a cut above and are pure min/maxers so I have to rely on luck to even get anywhere close!

 

[Haywire]

Bravo! Well researched and detailed. Time to re-install D2. Found Jah in Travincal many moons ago off some lowly caster.

 

[CRG]

This is fantastic, and it confirms why I'll never enter the untwinked 99'er thread - too much time to spend finding Bers for an Infinity, a runeword I don't need two of (same re: Enigma; CtA (mine is +6 BO)).

 

[Codoh1]

Can anyone tell me how I recognize the River of Flame superchests?

 

[Pb_pal1]

Can anyone tell me how I recognize the River of Flame superchests?

Take a look here for explanations and even a nice picture of the formation you'll want to look for. Unlike LK, it's either there or it's not, which makes it easier to look for in that respect.

 

[Brak]

So I decided to go back and look at the Sur p78 LK patterns in more detail. I am about 95% sure that at least one pattern represents a locked version of another pattern. Unfortunately there are no screenshot records for 5 of the 11 patterns.

But if we look at the reported pattern for #7 it is
Amulet, Pike, Ware Axe and 2 others
Furthermore pattern #7 is only reported once, whereas most other range from 2-5 confirmations.

Now pattern #11 was found twice and has a cleaned screenshot that includes
Lochaber Axe, Pike and War Axe
I would posit this is the unlocked version of #7

The only other one that stood out was pattern #5 that was reported even without a screenshot and only once. It is Martel de Fer, Swirling Crystal. Given only two items, it is a bit less likely to exist as a locked only variant, though.

But I would then downgrade LK to 10 Sur patterns instead of the accepted 11. Though this doesn't really change a whole lot.

 

[mrriddler]

Very nice. This will be very helpful for me!

Though one...math issue...sorry. I think I encountered a similar issue in another game.

I'm almost certain the run average for pul+/vex+ is way too high to be true for the LK runes. It looks like you got those numbers from (65536/(82*6), but that can't be right. This equation assumes something like where a particular pattern seed is removed from the pool after it appears until all 65536 patterns are generated.

The calculation and workup I found is below and I *think*/hope it's right as algebra was a long time ago. Had to use a spreadsheet as it had two variables.

65536 total patterns
82 patterns for pul+
27 patterns for vex+
6 chests per run

We want to calculate the probability for at least (ONE) pul+ over an unknown number of runs. It's easiest to setup equation to find the probability for no pul+ drop per chest/run first.

No pul+ per chest = [(65536-82)/65536]
No Pul+ per map/run = [(65536-82)/65536]^6 = 0.992516119939662 (So 1-0.992516119939662 gives probability of pul+ find in one run which logically seems to check out. It should be very low.)

x= # of runs
p=probability of at LEAST one pul or better rune over x runs
p = 1-(0.992516119939662)^x

This took me a LONG time til I realized what I needed to do. Take the probibilty of NO pul and raise it to the power x(runs) to give the probability of no pul (or better runes) over the course of all runs. Then just minus it out of 1 and you get the probability of at LEAST one pul or better rune in the course of ALL those runs.

50% probability of pul+ = 93 runs
95% probability of pul+ = 399 runs

For vex+ (27 patterns) numbers. The equation is
p=1-(0.997530620776834)^x
50% probability of vex+ = 281
95% probability of vex+ = 1212

Numbers look somewhat in line with my experiences. It does get ridiculous if you consider specific high runs.

For ber ONLY (3 patterns)
50% probability of ber (at least one) = 2524 runs
edit: 95% probabiltity of ber = 10907 runs !!! (well it's what my numbers said but ahm yeah seems insane here)

Yep 2500+ runs for a 50% chance to get ber. I never got one from LK but maybe others can share anacdotal info?

One problem (big or not I'm not sure) even assuming everything above is correct. The problem with above setup is that it also assumes that each of the 6 chests on a map can generate ANY of the 65536 patterns and I'm not sure if that's actually the case. Anyone know if the same (chest pattern) seed can spawn multiple times on the same map/run?

Edit: Oh, I used 50% to estimate "average" run number needed and 95% to give an approximate value for runs needed to have *almost* guarantee chance of target rune(s). Was gonna use 99% but the run number gets ridiculous.

Edit2: Just realized the numbers I crunched indicate a faster/better return than yours Gripphon, which is opposite of what I thought...well at least the 50% run number is way lower. Wow...whoops. I'll have to recheck my work after some sleep or maybe someone else can do it for me?

Edit3: Wow, I think I confused myself. My average SHOULD be lower than what you had Gripphon, but the high probability run calculation (for 95% chance of find) should be much higher.

 

[Gripphon1]

@downding

I used mean of binomial distribution to calculate expected value of positive outcomes. We know what is the probability per opened chest and then we can simply know how many chests should we open at average to see some rune. Same method is used to calculate expected number of heads after X coin tosses.

Your way is good for calculating the odds of something to happen, but 50% chance is not connected with expected value as you have put it. To demonstrate this I'll use simple coin toss twice in a row. Let's see what are the odds to get at least one head in 2 coin tosses:

0.5^2 = 0.25
Chance to get at least one head: 1-0.25 = 0.75 = 75%
What is the expected number of heads we could see in 2 coin tosses if we do the experiment infinite number of times? 2 tries * 0.5 probability = 1. Kinda expected. Or ,we are expected to find one head in 2 coin tosses where we also have 75% chance to find at least one head.

If we assume Vex+ does drop once in 405 runs aka once in 405*6 = 2430 chests when finding Vex+ has probability of 27/65536 per chest?
1 - [(65536-27)/65536]^2430 = 0.6326
Or, we have 63.6% chance to find Vex+ in 2430 opened chests while Vex+ at average drops once in 2430 opened chests as per mean of binomial distribution. It doesn't exclude the fact pattern could repeat itself.

This is interesting thing about probabilities, when doing such calculation, odds of successful outcome when we do number of tries equal to expected value, then odds are closing on to ~63% probability to succeed, not ~50%. With coin toss value is at 75% instead, but the more tries we do, the more it closes up to 63%. Why is that, I don't know. At first I also thought 50% chance could be used to determine expected number of tries if we multiply number of runs by 2 same as we do with coin tosses, but it doesn't work that way for some reason.

As for personal experience on such odds, that is pretty inaccurate because odds are way too big for our personal experience to be much of use in that sense, only as some general guideline. We need someone do a lot of running for results to have any value of comparison. For example, I did total of 140 hours of Travincal running on tournaments, and my total runefinds look like this:

Cham: 5
Jah: 8
Ber: 1
Sur: 4
Lo: 6
Ohm: 13
Vex: 10
Gul: 20
Ist: 19
Mal: 17
Um: 18

Ber+ count: 14
Vex+ count: 47
Um-Gul count: 74
Um+ count: 135

While it is obvious that some runes are easier to find than the others, it is still not precise to calculate anything. We know drop odds of runes individually and we could calculate some expectations. For this purpose I will use my presented numbers for Travincal p3 and will have to estimate average runtime since I also ran with barb some runs, but I know his numbers and could translate it to sorceress runs. In those 140 hours I in total did ~27465 runs (LOL) with sorceress on p3 with average time ~18.35 seconds. Expected number of finds should be this:

Um-Cham: ~ 237 runs
Vex-Cham: ~ 572 runs
Ber-Cham: ~ 1934 runs
(these values are from binomial distribution same I calculated for LK)

My runs suggest this

Um-Cham: ~ 203 runs
Vex-Cham: ~ 584 runs
Ber-Cham: ~ 1962 runs

Reasonably close, but still not perfect obviously. If I go to compare into lower odds, then results don't make sense anymore. My results suggest that Cham is 5 times easier to find than Ber. In fact Jah is almost as easy to find as Vex. Gul is easiest to find of them all... Normally, when we go into too low odds, then we get such ridiculous results. Vex+ and such values are much higher odds so they actually make sense.

Even my 140 hours is probably not good enough, but I will do more runs in the future and could provide even more precise data on everything, or at least I expect I won't have periods of mad luck or poor luck in front of me. Values should become closer and closer to expected values I presented.

 

[Nano]

This is interesting thing about probabilities, when doing such calculation, odds of successful outcome when we do number of tries equal to expected value, then odds are closing on to ~63% probability to succeed, not ~50%. With coin toss value is at 75% instead, but the more tries we do, the more it closes up to 63%. Why is that, I don't know. At first I also thought 50% chance could be used to determine expected number of tries if we multiply number of runs by 2 same as we do with coin tosses, but it doesn't work that way for some reason.

Just wanted to expand on this, since I find this interesting. If a drop has a 1 in N chance of occurring and is attempted N times, the chances of it not dropping would be (1-1/N)^N. As N approaches infinity, that expression rapidly converges to 1/e, where e is the natural number (2.71828...etc). So the chances of the drop occurring at least once approaches 1 - 1/e, or 63.2%, as N increases.

Anyways, I agree with the use of expected values as Gripphon presented, rather than trying to find the probability of at least 1 pattern or not finding a pattern. The latter approach can give deceiving numbers, such as:

Numbers look somewhat in line with my experiences. It does get ridiculous if you consider specific high runs.

For ber ONLY (3 patterns)
50% probability of ber (at least one) = 2524 runs
edit: 95% probabiltity of ber = 10907 runs !!! (well it's what my numbers said but ahm yeah seems insane here)

Yep 2500+ runs for a 50% chance to get ber. I never got one from LK but maybe others can share anacdotal info?

The more accurate statement would be 2500+ runs for a 50% chance to get at least one ber (pattern). Or 95% chance for at least one ber in 10907 runs (expected number of ber patterns is almost 3 for this many runs). This doesn't account for the chance of multiple ber rune drops. I feel this approach is more suitable for something that is only needed to occur once, like a lottery for example. If you can profit from multiple drops, expected value approach gives a better representation.

I'm almost certain the run average for pul+/vex+ is way too high to be true for the LK runes. It looks like you got those numbers from (65536/(82*6), but that can't be right. This equation assumes something like where a particular pattern seed is removed from the pool after it appears until all 65536 patterns are generated.

Binomial distribution uses statistically independent events, so nothing is removed from the pool.

 

[mrriddler]

Gripphon and Nanomist. Thanks! That was exactly what I got wrong. It's not 50% but 63% probability I should be using to calculate for the expected run number. In that case, I get the exact same result as Gripphon had originally.

p (probability for pul+) = 1-(0.992516119939662)^x

x (runs) probability
132 0.629012634993
133 0.631789059936 ~ 0.632
134 0.634544706449

I guess my followup question is what sort of error calculation could be used to guage the variability from that expected number? Would there be an "easy" way to do that like a standard deviation or some other function? Like for example say I ran 1000 LK without a vex+ when I could have expected at least one vex or better in around 400 runs. How should I be interpreting my result...etc.

Welp, time to farm some trav in 1.13 then. I'll hit LK again in a bit once I take some break. Running racks from 1.07 is absolutely brutal.

 

[Gripphon1]

I don't think you need any special variability, you can only get headache from it. I think things even out nicely in diablo after good amount of running (5k+ probably) while variability on only 1k runs is too high.

 

[abxl]

Hey i've got 2 questions if anyone is still there by any chance

How do you borrow a map from another character ? Just copy the .map file from save folder or anything else ?
What are these 30+ max dmg gcs and 10 max dmg scs youre using ? 1.07 patch huh ?

 

[ffs]

@abxl Borrowing maps from other characters is against forum rules. A few changes to SPF rules are currently being discussed including this issue, but the vast majority of users so far reject any substantial change to this particular rule. You can find some information on that in the discussion threads linked in the survey thread if you want to know anyway.

Just to clarify: The use of another character's map by @Gripphon for the purposes of this guide was purely for testing reasons to find out how different builds compare running Travincal under ideal (i.e. "perfect map") circumstances. Any progress, including, XP gained and runes and any other items found in such runs, was subsequently reversed/deleted.

And yes those charms are from 1.07 or 1.08 AFAIK.

 

[abxl]

i'm not really taking part in any tournaments, i'm just playing single player using <removed> for my own joy and the purpose of saving a map is mostly because it happened to me in the past that I accidently missclicked and changed my difficulty so the map was lost forever ^^ if you don't want to share the information about that, it's ok I guess, i'll do my best to search the mentioned threads out, thanks for the reply though ^^