A little theory on the RNG

[Lunariel]

I've been doing tailoring with mostly Ceremonial Solstice Robes since Wu's. My experience so far is that minimum and maximum success rates are not 5% and 95%. When I started on the robes I expected to succeed on about one in twenty, but it was more like one in thirty to forty.

Also the trivial subcombines for the robe seems to succeed more than 95%. A rate of 95% would imply that I would lose one item for every stack i combine. After preparing each batch for my robes, I would count how much I had left. I would normally be at 96-97% and occasionally in the 93-99% range.

I have absolutely no doubt the programmers coded in 5% and 95% as the limits, just like with prices and other things. I think the real culprit is the Random Number Generator.

They have admitted that the RNG is designed to be streaky. And all of us tradeskillers have certainly noticed it. This means the next random value is weighted against the previous ones. I suspect this clumps the long run distribution of values slightly away from the extreme ends.

In a way it's just like a d20 die that has less probability for 1 and 20 than the other values. If this is true, it makes a lot of other things like skillups harder then they look on paper. And a rare drop flagged as one in fifty, would be even rarer than that.

Lunariel

[Ilona]

I was averaging one in 26 combines on solstice robes before hitting 200 skill.

[Iannyen]

Someone out there check me on this, but the 5% success chance only applies to a single combine attempt, right? So when you begin to look to multiple attempts that test individually, regardless of any previous attempts, then the 5% chance would begin to decrease, correct? It would still be a 5% chance to succeed each robe, not a 1 in 20 success rate, right? Am I makin sense, or is this just jibba jabba?

I dont' know if this helps, but 5% seems fairly close for me in brewing.

For making Grobb meat last night for instance...

I bought enough ingredients to make 240 of them.

Ended up with 216.

Thats about average, and equates to very close to 5% (trivial).

If you also consider that I failed on 7 of the 140 sub combines, so I only had 37 out of 40 final combines, and failed on one of them.

More specifically I did the worst on Rat meat I think it was, and only had 37 of them, so I ended up destroying my 2 extra of the other meat, 2 extra soda water, etc to clean up the kitchen as it were.

[Liwsa]

Del,

Why didn't ya just make 2 more rats??

[Wandor]

Re: A little theory on the RNG

This means the next random value is weighted against the previous ones.

You realize that even if this where true, the same RNG is used by everyone. In the time that it takes you to click combine, load up ingredients again, and click combine again, the RNG has cycled hundereds if not thousands of times to other players across the server.

Why didn't ya just make 2 more rats??

Few reasons, I combine in Erudin... old habit, bound at the still there that no one else in the world uses. Its home, and I only need 100 cha to get best price there, so I don't have to switch gear between brewing/buying from local merchants. Buy casks/water right there.

I also do my brewing before I go to bed, and I am not about to go back to shadowhaven to buy 2 of everything to even things out for 2 more grobb attempts.

Maybe if it was more lopsided, destroying 5 or 6 of one I probably wouldn't do. Also don't like leaving 1-2 in bank, because next time I'm buying full stacks anyhow.

Long answer to a short question, sorry.

[Lilosh]

Someone out there check me on this, but the 5% success chance only applies to a single combine attempt, right? So when you begin to look to multiple attempts that test individually, regardless of any previous attempts, then the 5% chance would begin to decrease, correct? It would still be a 5% chance to succeed each robe, not a 1 in 20 success rate, right? Am I makin sense, or is this just jibba jabba?

Jibba Jabba.
---------------------------------------------------------------
When I roll a die, there is a 1 in 6 chance that I roll a 6.

This means that, in a perfect world, If I roll the dice 6 times, I get a single 6.

If I roll a die 60 times, I will get 10 6's.

If I roll a die 600 times, I will get 100 6's.
-----------------------------------------------------------------

A 5% chance of it happening on one occurance means it will happen 5% of the time.


In a perfect world, with a perfect RNG.


-Lilosh

[sir0die]

Re: A little theory on the RNG

This means the next random value is weighted against the previous ones.

You realize that even if this where true, the same RNG is used by everyone. In the time that it takes you to click combine, load up ingredients again, and click combine again, the RNG has cycled hundereds if not thousands of times to other players across the server.

ok, i was under the impression that each zone is hosted by it's own server. When you say for example the Inny server, actually refers to several computers (1 per zone) that collectively make up the Inny "server". Therefore the RNG is actually different depending on the zone you are in.

if you are the only player in a zone, you make your item and by the time you load and click combine, no one else has proccessed the RNGm, so you are getting the very next rng generated. If you are in a very populated zone (bazaar), by the time you click combine and reload an combine again, about 20 people have rng in the casino, 30 other tradeskillers just succeeded/failed their combines, so you are effectively getting the 1st and the 51st rng.

I think a better way to explain anomolies is this...

The chances of getting 'Heads' 100 times in a row flipping a quarter are astronomical.

The chance of the 101st flip landing on heads is still 50/50.

[Lunariel]

The next random seed and possible weighted data can either be stored for each player, each zone or each zone server. It shouldn't matter for this discussion. In the long run, the distribution of values should be the same.

I've made over 3500 blessed dusts so far (pottery above 180), and slightly less blessed silk. In the beginning I did smaller batches and it was hard to tell any pattern. But lately I've been doing batches of up to 300 dusts and then around 290 silks. Most of them ended up well below 5% failure, except once where I had one failure more than 5%. Sadly I have no exact count, but I would guess an average of just under 4%.

It may be that I've been just lucky with those combines, but my intuition tells me the RNG doesn't have a completely even distribution of the values. That's why I propose the theory that the extreme values are less likely. Feel free to tear it down with reports of 5% failure.

Lunariel

[Iannyen1]

I think a better way to explain anomolies is this...

The chances of getting 'Heads' 100 times in a row flipping a quarter are astronomical.

The chance of the 101st flip landing on heads is still 50/50.

Thats what I was trying to get at. Thanks.

[Emelyn]

The following is just the speculation of one humble programmer.

EQ uses so many random numbers every second of the game, consider how many variables in a simple one on one combat with a level one skeleton. Did you hit or miss, if you hit how much for, did it hit or miss, etc.

A) Given the vast number of random numbers used in the game
B) knowing that any good random number generator program is fairly computationally expensive
C) assuming that EQ tries to optimize for best performance so their servers aren't overloaded - with all that the servers have to do I don't think they want to invest a large portion of that into random numbers
D) knowing that less computationally expensive random generator algorithms are generally very very poor

Given all the above, I have long ago surmised that EQ does not use a random number generator algorithm at all. They probably just use streams of data (heh, the byte code from a compiled program for example). Or they may generate a long series of random numbers during server up, low usage periods, or something like that, that they then cache and reuse over and over.

I could be wrong, but that's how I would do it if someone wanted a cheap efficient random number scheme.