It's Past Time For An Rng Rework: My Suggestion

[Mr_Zurkon]

We've all complained about RNG at one time or another. I wanted to be able to make a constructive suggestion as to how to solve the issue, which is why I've spent a fair amount of time mulling this over before posting about it. Fair warning for those who are squeamish about math: there will be some talk in here about discrete probability below.

 

 

Now, I don't believe the problem with RNG is the size of the drop tables, or necessarily the probability weights assigned to the items within them. I think the real problem is with the concept of randomness. The important thing to keep in mind here is that the expected value of a discrete random variable over a discrete probability distribution is only guaranteed to match reality over a number of samples that approaches infinite (the reverse of the law of large numbers). Plotting the outcomes of those samples for any single player is unlikely to correlate well at all. And the experience of the individual player is what really counts when measuring whether or not RNG is doing a good job. In its current state, the grind is very inconsistent from player to player. Some very lucky people get what they want very quickly, and some very unlucky people will run missions tens or hundreds of times without getting the MacGuffin they're after.

 

Here's an example using assumed probability values for the Tower I Capture data tables:

 

Odonata Prime Blueprint -  0.0625

Paris Prime Grip - 0.0625

Forma Blueprint - 0.0625

Tower Key - 0.25

Boar Prime Receiver - 0.25

Lex Prime Receiver - 0.0625

Tower II Capture Key - 0.25

 

Looking at this, I can see that I actually told a small lie above when I said the problem isn't the drop tables, at least in this case, as you have a ~52% chance of getting some sort of tower key (assuming this data is accurate, and I have no reason to doubt it at this time). Anyway, let's do the math and see what our expected value is going to be for a given run of Tower I Capture.

 

Parametrizing the possible outcomes like so (which is not perfect, but it gets the idea across, I thin):

Tower Key = 1

Tower II Capture Key = 2

Boar Prime Receiver = 3

Paris Prime Grip = 4

Forma Blueprint = 5

Lex Prime Receiver = 6

Odonata Prime Blueprint = 7

 

µ = (1 * 0.25) + (2 * 0.25) + (3 * 0.25) + (4 * 0.0625) + (5 * 0.0625) + (6 * 0.0625) + (7 * 0.0625)

µ = 0.25 + 0.5 + 0.75 + 0.25 + 0.3125 + 0.375 + 0.4375

µ = 2.875

 

This indicates that the expected outcome for any given run of Tower I Capture is most likely to result in a Tower II Capture Key or a Boar Prime Receiver (which we could also tell from the fact that TowerKey_Pi + TowerIICaptureKey_Pi + BoarPrimeReceiver_Pi = 0.75)

 

So, the probability of getting the "good stuff" is sorely stacked against the player. Which is not necessarily a problem in itself, either, because it makes good sense for some of these things to actually be relatively rare. Here's where the problem comes in: remember back in the second paragraph when I mentioned the reverse law of large numbers? This is where our paint point exists - at the intersection of Probability Dr. and Reality Blvd.

 

Looking at the previously posted drop table, it is natural to intuit that, if you run Tower I Capture 10 times, you will probably end up with 5 keys, 2 boar prime receivers, and 3 of some selection of the remaining 4 possible rewards. Despite the probability of the thing, it is entirely possible to run the Tower I Capture 10 times and get 10 tower keys (and not even all that unlikely - the probability of getting 10 tower keys in a row is roughly 1.4%. Meanwhile, the odds that the Odonata Prime Blueprint drops at some point during the course of 10 runs is a mere 52%.)

 

----

 

Those of you still reading this may be wondering when I'll eventually get to the point. I don't blame you, so here we go.

 

What this all culminates to, is essentially that true randomness is not intuitive. Using a straight up, discrete random chance to pick drops from a reward table is going to result in a lot of frustrating failed attempts to get the rewards we, the players, actually want. (This is particularly noticeable when running endless missions, getting the same reward from the same rotation several times in a row gets frustrating quickly).

 

But how do other games solve this problem? Surely we don't all feel this kind of frustration in all games incorporating drop tables of one sort or another. As it turns out, using discrete random distributions is the crux of the issue - that is, the outcome of any given draw from the drop table is independent of the result of any other draws from the table.

 

Some other games seek to get around this by using a "fair" distribution instead. For example, the official Random Generator specification for Tetris saves you from the horror of potentially drawing 20 S and Z tetrominoes in a row by, instead of selecting each individual tetromino randomly, shuffling a sequence of all seven tetrominoes and dealing them in that shuffled order before generating the next sequence.

 

Now, this is not an ideal solution for this problem, either, as it is solving a similar but separate space (and I don't think players in general would like having a number of their future drops determined ahead of time).

 

Tetris actually has another possible solution that I think fits a lot better, though: enter the Tetris: The Grand Master Randomizer (or the TGM randomizer).

 

Here's how the TGM Randomizer works: in an effort to make successive identical draws less common, a history of the four most recent tetrominoes is kept. When the next tetromino is chosen, if it is not in that history, then the tetromino is dealt as normal. If the tetromino is in the history, however, it discards that result and draws again (up to 4 times with TGM1 and 6 times with TGM2 and later). If a result comes up that is not in the history, it is used. If, after 4 or 6 tries, the resulting draw is still in the history, it is used anyway. In the 6-try variation, this reduces the probability of drawing a tetromino that was drawn within the previous 4 draws (as compared to a random chance without retry) from ~65% to ~3.5%.

 

In short: adapting some variation of the TGM Randomizer to WarFrame would do a lot to ease the frustration we all feel from time to time about RNG. I don't necessarily care when it takes me several tries to get the uncommon or rare reward that I actually want, but I often do care a lot when I end up getting the same reward several times in a row. I also think that, in order for it to make a tangible impact, the recent reward history should be kept per table rather than globally across all rewards (because endless missions involve multiple drop tables, and we as players don't always run the same mission back to back when trying to grind out a given desired reward).

 

So, yeah. There's my idea. Thoughts?

Edited April 5, 2015 by Mr_Zurkon

[Genomega]

First of all remove dataminded data you can get into big trouble for that

[(PSN)WiiConquered]

I haven't read the post, but don't get it shut down for datamining. Instead just have estimates or made up numbers (which will make the math simpler anyway).

[Grom-84]

Looking at the previously posted drop table, it is natural to intuit that, if you run Tower I Capture 10 times, you will probably end up with 5 keys, 2 boar prime receivers, and 3 of some selection of the remaining 4 possible rewards. Despite the probability of the thing, it is entirely possible to run the Tower I Capture 10 times and get 10 tower keys (and not even all that unlikely - the probability of getting 10 tower keys in a row is roughly 1.4%. Meanwhile, the odds that the Odonata Prime Blueprint drops at some point during the course of 10 runs is a mere 44%.)

Yea that was my case yesterday - 10 runs 8 keys 2 formas - enjoy. But everyone seems happy, no changes are needed because they will ruin the game!)

Edited April 5, 2015 by Grom-84

[(PSN)WiiConquered]

I do like the idea now that I've read it, and I think others will as well. Even more reason to remove the datamined info so your post doesn't get shut down.

[Genomega]

Yea that was my case yesterday - 10 runs 8 keys 2 formas - enjoy. But everyone seems happy, no changes are needed because they will ruin the game!)

And the trading chat

[420degreequicksopeswag]

First of all remove dataminded data you can get into big trouble for that

Agreed, get it out of here QUICK.

[gnat6]

I do like your idea but those drop tables are a much bigger issue and any attempt to reduce the void grind is going to be pointless until DE decides to knock it off with the absurd drop chance weighting. 

[Wawazat]

just change the exact values to more clean numbers you "assumed" and remove any talk of data mined and name shame. your suggestions are nice and i think everyone would like them a lot

[Grom-84]

And the trading chat

And whats about it? nobody wants to buy that generic stuff everyone wants to buy/sell nova, volt, primed point blank, high voltage or maxed narrow minded.

[Gemini_Jenvas]

another problem is of course that they always make one component more rare than the rest, so the easy to get part clog up the drop tables.

Untamed rng is terrible, untamed rng on such roids is just plain aweful.

 

But, nothing will happen. DE makes money and their coding still sucks.

[Mr_Zurkon]

First of all remove dataminded data you can get into big trouble for that

Done - thanks for the warning!

[Mr_Zurkon]

But, nothing will happen. DE makes money and their coding still sucks.

 

Let's not devolve this thread into DE bashing, please. They have a strong history of listening to community feedback and acting upon it. I don't think this has anything to do with money - they have a lot of issues to tackle and can only do so much at once with the resources they have available.

 

I'd like to think that even if they don't use my idea, maybe it will set the brain cogs in motion and they will have their own idea to resolve the RNG issues. Either way, I would appreciate it if we could keep discussion in this thread productive.

Edited April 5, 2015 by Mr_Zurkon