Gradual Reduction in MF Effectiveness

[Jeger]

So if you get a double-durability set item, that has to be considered a set item for purposes of drop "goodness", since you would have gotten a set item if it was at all possible.

Not necessarily. The random number generator does not generate random numbers, but pseudo-random ones that are generated by making a call to a mathematical function that returns numbers that "behave randomly." These functions are sought after and tested.

The testing is difficult. For example, if function is used to generate a number between 1 and 10. One test would be to determine what the odds are that you would get a particular number. If it generates each number roughly 10% of the time - it passes that test. Suppose, though, it generated a 1 five percent of the time after it generated a 9, but 15% of the time after it generated an 8. The odds of getting a 1 on any call would still be 10%, but the RNG has a degree of dependency, and is not a good RNG.

It goes on. Suppose that after six "extreme" numbers (1, 2, 9 or 0) there is a tendency for a "middleish" one (3-8) ? That would be more difficult to find. Much effort is invested in testing these routines, and I will take it for granted that Blizzard chose a good one. However, it still is not random. This has always been a problem for games. People are very good at figuring out patterns. I played an arcade game where I started to "know" how the enemy would behave. I think I was seeing a failure in the RNG routine that resulted in a subtle pattern that my brain learned.

I don't know how many numbers it takes to generate an item - let's say it is four. Suppose, that the RNG tends to fall into a pattern after many calls, so there is less of a chance that a group of four numbers will result in a valid unique, but more of a chance that it will be a non-valid unique. In that case, the observation could be a RNG failure not a deliberate MF reduction.

This is one explanation, that may not hold water, but there are others.

[Obi2Kenobi]

If it has double [triple] durability, this comes after any Random number. It is only given double [triple] durability if there is no set [unique] item for that type of item (MF won't help you here), or if the monster can't drop it (MF won't help you here either). So it should be considered a "set" ["unique] drop.

[Jeger]

So it should be considered a "set" ["unique] drop.

No, because part of the problem could be fewer "real" uniques and more "triple durability" uniques.

[LiquidDamage]

No, because part of the problem could be fewer "real" uniques and more "triple durability" uniques.

You mean fewer proportionally, I assume?

There is already a massive "problem" with 87% of item types not having a set equivalent, and 31% of them not having a unique equivalent.

I seem to recall reading something back around 1.08 about how the xpack-only class-specific items were added to the TCs in such a way that you get a disproportionately large amount of class-specific items when compared to the total number of items. In other words, if there are, say, 50 class specific items, instead of 1/10 of all items dropped being class specific, 1/5 are, or something along those lines.

If this thing I'm remembering isn't total crap (I'm sure Ruvanal or adeyke would remember this if it is true), then since there is only one class-specific set item per class, and 3-4 class specific uniques per class, then indeed, a disproportionately large number of failed sets and uniques are dropping, but through no fault of the RNG.

Edit: I reread your post and see that you were suggesting that as time went on, the RNG would tend to favor failed sets and uniques, as opposed to it just favoring them entirely. So, my response doesn't apply to what you are saying, but I will leave it anyway, as I am curious to see if the bit about class-specific items being more common is true.

[Jeger]

1-4: 3u 2s 7r 0m
5-8: 2u 3s 7r 5m
9-12: 0u 2s 6r 3m
13-16: 2u 5s 7r 3m

The drops dip in the 3rd run, but it doesn't seem like much to me. Make what you want of it.

Well, since this is the only actual data we have, let me over-analyze it:

Uniques

Chance that the four sets should be ordered with the highest unique value in the first two entries, and the lowest unique value in the second:

(1/2) * (2/3) = 1/3

so if it were truly random, the data would look this bad 33% of the time. Not significant, but interesting.


Uniques and Sets

10 uniques and sets in the first group, 9 in the second. Chance that this occurs is exactly 50%. Not at all significant.

Obviously, we need more data. :P

[adeyke]

I'm not sure I understand what you're talking about.

The procedure for picking an item is this:
- Pick a base item (there's no knowledge of which items have sets/uniques here).
- Pick a quality for the item (there's no knowledge of which items have sets/uniques here).
- If the quality is unique [set], pick from the possible uniques [set items] for that base item. Else, make it a triple durability rare [double durability magic].
- If the specific unique that was picked was already spawned in that game, make it a triple durability rare.

Now, if you're saying that part of the supposed worsening of drops is that it picks more base items that don't have a unique/set equivalent, you're wrong, since when the base item is picked from a TC, the uniques and sets don't factor into it at all.

Consider two items:
Item 1: green swirling crystal (Tal Rasha)
Item 2: blue double-durability magic dimensional shard

For item 1, it first picked swirling crystal as base item, then set as quality. There's only one set swirling crystal, so it picked Tal Rasha.

For item 2, it first picked dimensional shard as base item, then set as quality. There's no set dimensional shard, so it had to be a blue double-durability item instead.

If you're counting the number of set items, both of these would have to count, since it had no know way of knowing that there are no set dimensional shards when it picked that base item and that quality.

[Jeger]

I'm not sure I understand what you're talking about... ... (there's no knowledge of which items have sets/uniques here)

I'm talking about a failure in the random number generator. There need not be any knowledge of which items have sets/uniques. It would be a result of a co-variance between successive random numbers that causes certain combinations of numbers to be more likely than others as time goes on. If the tables were arranged in a way so these "more likely" numbers resulted in high-durability magic items, it could explain what people see. It's not really what I think is happening, so it's probably not worth going into in depth. What it means is that counting high-durability magic items as uniques will not properly test all possibilities.

Pardon the trip into excessive nerdiness here, but there is a simple test for one-degree of co-variance that might be fun to run on the Blizzard RNG. [Peers through his thick glasses at the monitor.] Start with a 600 by 600 pixel bitmap with all pixels colored red. Go into a loop where you pick a random X (1-600) and random Y (1-600.) Examine that pixel and turn it black if it is red. If it is not red, make it slightly whiter. Let it run.

What you should see is a red graph with points rapidly turning black. As time goes on, the red points should disappear, and the screen should gradually turn white as the points are hit again and again. Let it run "a long time."

Eventually all the red points should disappear. Examine the graph to see if there are suspicious darker areas or lighter areas. If there are, make note of where they are. Run the program again and see if they are still there in the same place. If so, the RNG fails because it has one-degree of co-variance. It also fails if some points stay red forever. Blizzard's RNG should easily pass this test, though. If not, I am very disappointed in Blizzard. Good RNGs will pass that test easily.

[gekko]

I'm talking about a failure in the random number generator. There need not be any knowledge of which items have sets/uniques. It would be a result of a co-variance between successive random numbers that causes certain combinations of numbers to be more likely than others as time goes on.

This still doesn't apply here.

We're discussing whether or not MF decreases in effectiveness over time. For the purpose of deciding how often the quality level of an item is chosen to be "unique" or "set," those double or tripple blue or yellow items *were chosen to be unique or set.* Let me repeat that. If the final few steps were somehow bypassed or ignored, those items were not blues or yellows -- they were sets or uniques. The step of changing their quality from unique or set to magic or rare has nothing the least bit random about it -- if one of very specific conditions are/are not met, the item spawns as an increased durability magic/rare item.

Consider this: roll a 6 sided die 100 times. If you roll a 5, change it to a 1. Now count the number of times you rolled 1, 2, 3 or 4 compared to how many times you rolled 5 or 6. BUT count the 5's you changed to 1's as 1's. In other words, you will have zero 5's, and (roughly) twice as many 1's as any other number. According to the stats you just collected, your die must be weighted somewhere -- you'll have rolled 5 or 6 only about 1/6 of the time (assuming equal distribution and such). If you had counted the numbers you changed as 5's, you should record roughly a 2/6 chance of rolling a 5 or a 6.

While it IS possible to have a failure in the random number generator (after all, how can a computer choose a totally random number... for that matter, how can a human choose a totally random number?), that doesn't change the fact that increased durability blue/yellow items must be counted as unique/set drops for the numbers to be meaningfull. Because as far as the game is concerned, they WERE chosen to be unique/set.

gekko

[Jarulf]

It would be a result of a co-variance between successive random numbers that causes certain combinations of numbers to be more likely than others as time goes on.

Ehh, no, it would not change over time. It would always be more (or less) likely. Otherwise some other change would have taken place. So shifts (for runs or long playing) in ammount of failed uniques or sets can't be attributed to the prng no matter what as long as other outside parameters are the same (which they are if you kill same monster).

[twentythree]

So you roll a die 100 times and it comes up a 5 or a 6 only 1 out of every 6 times.

Then roll a die 100 more times. This next time you do it a 5 or a 6 only comes up 1 out of every 10 times.

Then roll it 100 more times. Now a 5 or a 6 is coming up only 1 out of 1very 20 times.

If it is true that a gradual slide off is happening this slide off will show up if you count only the 6es or if you count both the 6es and the 5-changed-to1-es. Likewise, if the magic find percentage is actually dropping then it doesn't matter if you count only uniques or if you count uniques+uberdurable. The same gradual slide off will show up under either system of counting. Now if the magic find percentage is not going down but instead the failed unique percentage is going up then counting the uberdurables would be needed. But this is not what people are witnessing so I don't see the need to count this way.

And besides, nobody does runs in hopes of getting high durability items. People WANT uniques. When or if anyone ever tests this and if they find that runs 1-10 generated a dozen uniques and runs 11-100 generated zero uniques then guess what? I will only do runs in batches of 10. You may try telling me "but runs 11-100 generated nine dozen items that were ~almost~ uniques so mathematically that is the same". I still wont do runs 11-100 if they wont generate uniques for me.

[Jarulf]

>So you roll a die 100 times and it comes up a 5 or a 6
>only 1 out of every 6 times.


So basically you got it 17 times.


>Then roll a die 100 more times. This next time you do it
>a 5 or a 6 only comes up 1 out of every 10 times.

So now, only 10 times.


>Then roll it 100 more times. Now a 5 or a 6 is coming up
>only 1 out of 1very 20 times.

And only 5 times. But what if it came out as a 1 (changed from 5), 5 times now but only 1 time above. The your "decrease" does in fact not exist and you are drawing conclusions on the wrong data. For the game you based it not on the number of times the game wanted to drop a unique, but rather on how many was failed or not. VERY different thing. You might as well base it on the number of chickens you saw in town. Well, not really, but almost. And I think that in many cases, this is what people might have seen or other similar effects.


>If it is true that a gradual slide off is happening this
>slide off will show up if you count only the 6es or if you
>count both the 6es and the 5-changed-to1-es.

But you need a LOT more samples then. And since people usually do very few (comparable), it is MUCH harder to draw conclusions if you do not count all.

[Jeger]

We're discussing whether or not MF decreases in effectiveness over time.

Actually, I believe we are discussing whether the treasure you get from repeated runs declines over time. If it does, one explanation would be that your MF declines, but there exist many others.

If we are strictly restricting ourselves to whether the MF declines over time, then it is true - you can count HDMI (high-durability magic items) as uniques. However, if we are discussing whether the treasure you get from repeated runs declines over time, then you can not count HDMI as uniques, because they are not. There exist many explanations (bad RNG is only one) why the ratio of HDMI to uniques could vary with time.

[Jeger]

And only 5 times. But what if it came out as a 1 (changed from 5), 5 times now but only 1 time above. The your "decrease" does in fact not exist and you are drawing conclusions on the wrong data. For the game you based it not on the number of times the game wanted to drop a unique, but rather on how many was failed or not. VERY different thing.

But this is what people ARE interested in. Not how many times the game "tried" to drop a unique, but how many times it DID. As I said before, many explanations exist that could account for the ratio of uniques to HDMI (high-durability magic items) changing over time.

[adeyke]

So your argument is that, due to some freakish flaw in the RNG, certain base items will occur less frequently over time and that these happen to coincide with the base items that don't have a set/unique version associated with it.

Such an argument is completely illogical.

Even if we were to assume that certain base items grow less common over time due to a bad RNG, there's no reason why it would coincide with the items that have set/unique versions.

TC armo87 has these items: Sacred Armor, Colossus Girdle, Ogre Gauntlets, Myrmidon Boots, Corona, Diadem, Guardian Crown, Dream Spirit, Vortex Shield, Bloodlord Skull. For simplicity, let's examine only the first six items (since none of those are class-specific).

Sacred Armour has a set version and two unique versions.
Colossus Girdle has neither set version nor unique version.
Ogre Gauntlets has a unique version, but no set version.
Myrmiddon Greaves has a unique version, but no set version.
Corona has a set version and a unique version.
Diadem has a set version and a unique version.

So, by your argument, Colossus Girdle (and perhaps Ogre Gauntlets and Myrmiddon Greaves) could become more common than the other items, so that, over time, you end up with more HDRIs and HDMIs than originally, and that this is the cause for a decrease number of uniques and set items.

Why?

If we assume that certain items would become more common than others, it could be just as possible that, say, Sacred Armour and Corona are the more common ones, or any other combination of items. Why would it happen to pick exactly those items that result in more HDRIs and HDMIs?

Here, we just looked at part of a single TC. There are hundreds of different base items. If the items that become more common are decided arbitrarily (that is, without looking at what sets/uniques are available), the chances that the more common ones happen to be the ones without set/unique versions would be really, really small.

And this was all with the assumption that base items that should be equally likely really aren't and that there's a trend towards some being more common as time goes on. There isn't any such effect that I know of.

Really, it doesn't make any sense.

[Jeger]

Really, it doesn't make any sense.

If you say so.

If you don't like that one, how about this one:

Blizzard as a compile switch on compilation that allows two different servers: one that serves single player and one that does battle net. 99.99% of the code is the same, and the same source code is used, so the programmer that talked to Jarulf was being honest. The BattleNet version keeps track of boss kills, and when your number gets too high, it does not reduce your MF, but instead starts to reroll the treasure a fraction of the time whenever it gets a unique in the loot.

If this were true, then the number of HDMIs would rise with time, and the actual uniques would decline. Therefore, you need to measure actual uniques, not just HDMIs.

As I said, there are any number of possible reasons that the ratio of actual uniques to HDMI could vary with time. If people are seeing a true drop off in treasure rate, this has to be allowed for.

To remind everyone: my current stance is that it is all psychological. I am just saying that if you really want to test it, you can not count HDMI as uniques, because they are not.

Why would it happen to pick exactly those items that result in more HDRIs and HDMIs?

If this is actually happening, it would be random luck - it could have gone the other way, then people would say loot increased with time.

So your argument is that, due to some freakish flaw in the RNG

It would not be freakish. All random number generators have flaws - the freakish ones are the ones whose flaws are so minor that you do not notice them.

Also, it's not really my argument. It's just one possible example of how the ratio of HDMI to uniques could change with time. There are many, many others (such as the one I gave above.) We could spend a lot of time arguing the merits of each one, but it would be pointless. The question won't be answered until someone takes the time to run a series of tests that is long enough to satisfy everyone. If they do, I'd like to see the uniques counted separately.

[LiquidDamage]

To remind everyone: my current stance is that it is all psychological. I am just saying that if you really want to test it, you can not count HDMI as uniques, because they are not.

When testing something, it makes sense to keep track of all the details. Often, you will set all variables constant but one, in order to see if there is a problem with that one variable.

If we are trying to find out if "good" items drop less over time, it makes sense to start by checking to see if MF decreases over time. If you are going to test whether MF is working properly, you must count high durability items as failed uniques/sets. This is because MF will actually cause more of these items, as well as uniques and sets, to appear. Given the massive number of item types that can drop, streakiness in the RNG might cause you to come to invalid conclusions, just because when you test it, it might happen to drop a disproportionate number of some item quality than it does for someone else, since you are ignoring some ~50% of "good" drops as they relate to magic find.

Think of it this way:

1) Joe does 200 runs which yield 10 uniques, 20 sets, and 10 failed uniques and 20 failed sets. He happens to get the 10 uniques and 20 sets early in the runs, and the failed ones late in the runs. He concludes that there is a dropoff in MF% as you do more runs, because he ignores the failed items.

2) Ten other people do 200 runs which yield the same number of items of each type, but they all actually note the durability of the items, and count them as failed uniques/sets. In checking their data, they find that, on average, their "good" drops still occur at the same rate regardless of how many runs they get.

This is why you must count them if measuring MF. You will be more likely to get inaccurate results if you don't.

Now, suppose the ten people in #2 consider your suggested situation: that more failed sets and uniques occur later in the runs. They now have the data to see if this is actually occurring, since they kept track of the high durability items.

In both cases, you need to pay attention to them for more accurate results.

[Jarulf]

There exist many explanations (bad RNG is only one) why the ratio of HDMI to uniques could vary with time.

No, since it doesn't have any time variable in it, that is not possible.

[Jeger]

If we are trying to find out if "good" items drop less over time, it makes sense to start by checking to see if MF decreases over time. If you are going to test whether MF is working properly, you must count high durability items as failed uniques/sets.

I agree. :)

If you are going to test whether MF is working properly, you must count high durability items as failed uniques/sets. However, if we are trying to find out if "good" items drop less over time, we must not count high durability items as "good" items.

[Jeger]

No, since it doesn't have any time variable in it, that is not possible.

Edit: I deleted my response, because this isn't really contributing to the thread any more, I think. People who want to discuss RNG theory can PM me. ;)

[LiquidDamage]

I agree.  :)

If you are going to test whether MF is working properly, you must count high durability items as failed uniques/sets. However, if we are trying to find out if "good" items drop less over time, we must not count high durability items as "good" items.

Sorry, this looks like a misunderstanding. When I put "good" in quotes, I mean that they SHOULD be good, from a quality perspective- i.e. did your MF actually make the drop better. I don't actually mean that the item is useful.

In other words, a normal durability rare Colossus Blade with 400% ED, dual leech, and 30% IAS would be a heck of a lot better than a Hellplague, but for determining MF effectiveness from a quest drop, the Hellplague is "good", and the Colossus Blade isn't.

The origin of this topic was about MF having reduced effectiveness over time, as the title says. I originally used "good" to mean that the item was supposed to be a set or unique. I wouldn't put it in quotes if I actually was referring to the item being good. We agree on all points (including the conclusion that the perceived reduction is psychological).