Ok, all you math geeks...

[tairne]

We've seen GREAT work applied to the trivial vs success formulas..

[code:1]Success* = Skill + 50 - Difficulty

Trivial = Difficulty + 16
Trivial = 4/3 * Difficulty + 2

Success = max(5,min(Success*,95))[/code:1]

What have you geniuses come up with for INT/WIS (STR/DEX) vs skillup?
How much and what kind of data do you need?

There is a new server opening up, a chance for a LOT of raw data from skill 0 on up will be available, soon. =)

[kiztent]

number of combines per skill up = (constant/relevant stat)

Where constant depends on a fudge factor for the level of the skill and the particular skill.

edit: wrong formula.

[zwobu]

Difficulty = 42, Trivial = 58.

Or is that an either/or situation on the Trivial-Difficulty formulae?

[Sylphan]

I don't know what an 'either-or situation' would be mathematically.

You use whichever Trivial-Difficulty formula yields the higher trivial.
max(first formula, second formula)

You've correctly found the turning point at 42. For difficulties under 42 the first formula is higher, for difficulties over 42 the second formula is higher.

[Sylphan]

Well... trivials are a simple deterministic formula, but skillup rates involve random numbers, probabilities, and thus statistics.

It will take large volumes of evidence to even get close to the skillup formulas, and then there will always be a margin of error. It's not nearly so easy. Plus there's the issues of successes and failures, geerlocks, and how close to trivial are the items you're combining.

We have so far determined that it doesn't help to have lots of high stats, you only need one high stat (int, wis, str, or whatever applies). We've also determined that it is worthwhile to spend a certain percentage of your budget on skillup-gear, because it'll save you money in the long run. ie the effect of stats on skill is not really tiny, like the effect of charisma on charms. It's believed that a) geerlocks have no direct effect on skillups, b) successes are more likely to result in a skillup, c) it doesn't matter how close the item is to trivial, as long as it's not trivial. But those three issues are still debated.

I'd like to find out if raising INT from 100 to 200 would double your skill rate or just raise it 50 percent or maybe only 20 percent. Maybe it would double your skillup rate when your skill is around 50, but only add 20 percent to skillup rate when your skill is around 200. I don't suppose we need to get much more accurate than that... even pinning it down to the nearest 20 percent would be very helpful.

[zwobu]

Either/or, mathematically speaking, would mean you either use one equation or the other. Two simultaneous equations in two variables begs a solution :twisted:

So, what we have here is a piecewise function then, where the equation depends on the part of the domain in which you are working.

Trivial = Difficulty + 16, Difficulty =< 42
Trivial = 4/3 * Difficulty + 2, Difficulty >= 42

[Boleslav]

I would assume that the effect of the 'driving stat' is the same, regardless of what stat is used. That is, high INT would be the same as high WIS, or high STR for smithing.

That said, I would run a test on an item that has a trivial so high that you can get to 100 or higher in skill without getting better than the 5% success rate. This would remove 'close to trivial' issues as a source of variation. Mino Hero Brew is perfect for this. It trivials at 248 and is cheap. Lots of clicks per combine through…

I would then make (use) 4 characters. I would pick a stat (INT or WIS) and set it at certain points for each toon. One toon would be at 50, the next 100, then 150, then 200 in the stat in question. Each character then would skill up from 0 to at least 100, recording how many attempts per skillup, how many success and failures, and whether each skillup came on a success or failure.

You would want a chart that looked something like this:
Skill attempts successes failures Skill on Success?
1 3 1 2 Y
2 5 0 5 N

If I had more time than sense I would then skill each character up to 150 or even 200. If I really had too much time on my hands I would repeat the experiment with new characters.

If we had data from two runs of this experiment on characters from skill 50 to 200 skilling up to 200 the math geeks (like me) could really make some conclusions about the effect of higher stats on skillup rate. We could also tell a lot about some other issues like getting skillups on a successful combine.

If I see the data come in I would be happy to analyze it…

Boleslav Forgehammer
Paladin of Brell in his 60th Campaign
E'ci – Destiny Awaits

[kiztent]

If we had data from two runs of this experiment on characters from skill 50 to 200 skilling up to 200 the math geeks (like me) could really make some conclusions about the effect of higher stats on skillup rate. We could also tell a lot about some other issues like getting skillups on a successful combine.

If I see the data come in I would be happy to analyze it…

Well... not really. There's a third unknown to consider: overall chance to skill up. Data would have to be compared across varying levels of relevant stats in the same skill range. Since we don't know the equation for skill and skill up chance, there's a good chance we will have bogus data creep into the mix.

[Boleslav]

Kiztent wrote:

Well... not really. There's a third unknown to consider: overall chance to skill up. Data would have to be compared across varying levels of relevant stats in the same skill range. Since we don't know the equation for skill and skill up chance, there's a good chance we will have bogus data creep into the mix.

Having toons at 50, 100, 150, and 200 in the stat qualifies as 'varying levels of relevant stats' to me. Keeping track of total combines, successes, and failures for each skill level preserves the data if we want to study what happens at any given skill range.

Certainly we cannot remove all possible confounding effects. However, the test I described would go a long way toward quantifying the effects of stats on skillup rates.

Was anything done on this with the old fish roll tests? It seems natural that they might have tried something like this back in the day.

Boleslav Forgehammer

[kiztent]

Well... not really. There's a third unknown to consider: overall chance to skill up. Data would have to be compared across varying levels of relevant stats in the same skill range. Since we don't know the equation for skill and skill up chance, there's a good chance we will have bogus data creep into the mix.

Having toons at 50, 100, 150, and 200 in the stat qualifies as 'varying levels of relevant stats' to me. Keeping track of total combines, successes, and failures for each skill level preserves the data if we want to study what happens at any given skill range.

Certainly we cannot remove all possible confounding effects. However, the test I described would go a long way toward quantifying the effects of stats on skillup rates.

Well, I'd agree, except, we're not trying to find a correlation. We already know there is one. We're trying to define the shape of a function. This function varies depending on skill level in some unknown way, as well as on relevant statistic in some unknown way. Since the actual numbers of combines per skill up are roughly random, unless there is a large concerted effort to gather data across both variables, I think we're pissing in the wind.