Fan Faire June 2005 Write-up (Plus Tanker handout)

[Bolas]

Really the problem is that we don't want it easier for others to catch up to us, we LIKED having a barrier from 160 to 190 that made it so that only the dedicated tradeskillers got to 200+. We don't want it easier for people to catch up to us. And we don't want it suddenly a lot harder to get the last few points of a tradeskill.

Please can't we have the old formula back? This is a nerf to those that have been long-time tradeskillers. Bleah.

-Bolas

[Xislaben]

If I remember correctly from C 10.0 tells it do to a floating point divide, it does not give a result limited to 1 decimal it keeps all the digits out to the limit of the floating point format. So the result kept would be 52.833333 or so. Close enough for the use to which it is being put.

Yes I believe that is accurate. Now multiply 52.833333 by 10, you get 528.33333. Now compare that to /random 1000. Anything after the original 10's place will be ignored, IE 528 = 528.33333 for the purposes of comparing it to the random integer number for pass 1. Granted it's a very very small %.

Were you to say multiply it by 100, and compare to /random 10000, you would have 5283.3333, giving you 3/10000ths more of a chance than the current method. It's miniscule, and honestly probably unimportant, but it's point shaving and only in the negative direction.

[Xislaben]

I've seen the code/algorithms for lots of random number generators, everything from "Quick and Dirty" to "Cryptographically strong" but I have never seen one who procuces floating point numbers, and I was wondering if that "truncating effect" could cause the intended/unintended side effect of lowering skillups dramatically?

Pass 2 formula: % = 12.5 - ((10 / 125.0) * (raw skill - 175))
Would yield:
- 293 = 3,06%

- 294 = 2,98%
....
- 299 = 2,58%

But if the random number generator returns a whole number from 1 to 100, you will never see it return 2.1 or 2.2 so at skill 294+ only 2 values out of 100 (or 2.00%) will get you past this check.

That would be a concern, yes, and for the same reason as pass 1, only on pass 2 it actually may have a significant and measurable detrimental effect. If a calculated 4.9% to make pass 2 really is a 4% chance due to comparing it to say /random 100, then you've lost out on 22.5% of what your calculated chance should have been. At lower percentages the difference becomes even more pronounced, as 2% and 2.98% have a 49% difference. If pass 2 is being mutiplied by 10 and compared to /random 1000 then it's the same issue as pass 1, that is only the first digit past the decimal is used for the purposes of comparison.

So at raw skill 294 is your pass 2 chance 2% (/rand 100), or 2.9% (/rand 1000), because it won't be 2.98% unless you're comparing it to /random 10000.

If pass 2 uses /random 100 then the problem is significant, if it's /random 1000 then it's not very significant.

Now, because we all like pretty pictures, I went ahead and made one comparing the new listed pass2 with what it really would be if compared to a /random 100 roll, it becomes a stairstep:


If however you're looking at the number multiplied by 10 and checked with /random 1000 you see a negligable difference:


In both cases above the yellow line is the listed steady progression that the new post 175 skilling is suposed to take for % chance to make pass #2. And the pink line is the real % chance to make pass #2 based on random rolls of 100 and 1000 respectively.

[Dyllwin]

I know it's too late to make this change, but a better way to handle the DoN patterns would have been to be able to summon a stack of generic patterns for your race, then have a subcomine (no-fail) that would create the specific pattern you needed. That way only one stack of each no-drop pattern for each level (expert, master etc) is needed at a time, and you create the pattern you need at creation time.

Or even better, use the generic pattern and a standard store-bought pattern in the recipe for the final piece instead of having the subcombine. That way there are no extra combines needed...

If they had the time they could make the change now, but it would take a good bit of time that could be spent doing other important things, so I doubt it would be feasible

[Tanker]

Good bits of speculation here. Looks like you've got all the bases covered, so I'll just point to the right one.

For the first pass, I take the percentage chance as a float, multiply it by 10 and make it an int, then random 1000. Mainly, this was to imitate the old method when I cleaned up the code while still using a percentage chance somewhere in the mix for readability. Essentially, the last "/ 10.0" didn't exist in the old formula, and it randomed 1000. I did the divide so you could get the answer as a percentage.

For pass 2, the old code would just cap your skill at 5 and 190 and roll 200 to try and get over it. This couldn't work anymore, so a real percentage is used. Both for symmetry with pass 1 and so there wouldn't be a loss of (almost) a whole percentage point, I used the same "multiply by 10 and random 1000" method. The clipping is <0.1%, and since the minimum chance to skillup in pass2 is 2.5%, you could possibly lose <4% of your chance to skillup at 299. Not completely trivial, but not noticeable given the possible ranges of combines needed for that skillup.

[Xislaben]

Good bits of speculation here. Looks like you've got all the bases covered, so I'll just point to the right one.

For the first pass, I take the percentage chance as a float, multiply it by 10 and make it an int, then random 1000. Mainly, this was to imitate the old method when I cleaned up the code while still using a percentage chance somewhere in the mix for readability. Essentially, the last "/ 10.0" didn't exist in the old formula, and it randomed 1000. I did the divide so you could get the answer as a percentage.

For pass 2, the old code would just cap your skill at 5 and 190 and roll 200 to try and get over it. This couldn't work anymore, so a real percentage is used. Both for symmetry with pass 1 and so there wouldn't be a loss of (almost) a whole percentage point, I used the same "multiply by 10 and random 1000" method. The clipping is <0.1%, and since the minimum chance to skillup in pass2 is 2.5%, you could possibly lose <4% of your chance to skillup at 299. Not completely trivial, but not noticeable given the possible ranges of combines needed for that skillup.

That's great news Tanker, thanks!

Then there's only one uncertainty left in order to perfectly replicate the percentage calculation, when you cast the (float*10) into an int, does it round or truncate? If it truncates there's no change from graph #2, if it rounds however it could round up, which would give miniscule but slightly better odds in some cases. Then I'd ask does 25.5 become 26 or 25, and be able to faithfully reproduce a calculator.

[Konjure]

Would like to start to thank Tanker for his detailed reply in how the calculations are performed.

Then there's only one uncertainty left in order to perfectly replicate the percentage calculation, when you cast the (float*10) into an int, does it round or truncate?

It will depend on programming language but I think I can answer this using my famous logic...
* I like Everquest.
* Tanker likes EverQuest.
* I also like "C", therefore Tanker must also like "C".
In C all typecast from float to int would result in truncating all decimals i.e. rounding down (unless it's a negative number then it would be rounded up.)

For those who doesn't trust my logic (and nobody should ) I think the following quote proves that he have done his math with truncation/clipping (rounding down) in mind, so presumeably that is what happens.

The clipping is <0.1%, and since the minimum chance to skillup in pass2 is 2.5%, you could possibly lose <4% of your chance to skillup at 299. Not completely trivial, but not noticeable given the possible ranges of combines needed for that skillup.

[zumbenn]

Would like to start to thank Tanker for his detailed reply in how the calculations are performed.



It will depend on programming language but I think I can answer this using my famous logic...
* I like Everquest.
* Tanker likes EverQuest.
* I also like "C", therefore Tanker must also like "C".
In C all typecast from float to int would result in truncating all decimals i.e. rounding down (unless it's a negative number then it would be rounded up.)

For those who doesn't trust my logic (and nobody should ) I think the following quote proves that he have done his math with truncation/clipping (rounding down) in mind, so presumeably that is what happens.

It terms of speed, truncating was faster than rounding in the olden days. With modern chips, both operations are the same speed. BUT it depends on the C compiler being used. For sure, truncating in C is translated to a direct CPU instruction. Rounding in C may be a function call (slow) or optimized to a CPU instruction (fast).

/em Steps off his geek soapbox

Maevenniia, I ran into you in WoS yesterday while I was playing my chanter, Rosepedle (and making the comment about your name). Actually, it was almost impossible to see your real name because of the new "shrink long names" code. It was cool to run into someone from the boards tho.

[Bobaten]

Since mastery checks occur before the minimum fail check, does this mean that most of the mastery AA's are useless at 300 skill?
There seems to be a consensus that the highest trivial tradeskill items (grandmaster gear, and OOW augments) are using the optional minimum fail code. The failure rate that people seem to be reporting is between 20-30% (this certainly holds true for my augment cutting). I had been planning on picking up JCM to up my success rates, but now it appears that it won't actually give me any benefit.

[Aeght]

Since mastery checks occur before the minimum fail check, does this mean that most of the mastery AA's are useless at 300 skill?
There seems to be a consensus that the highest trivial tradeskill items (grandmaster gear, and OOW augments) are using the optional minimum fail code. The failure rate that people seem to be reporting is between 20-30% (this certainly holds true for my augment cutting). I had been planning on picking up JCM to up my success rates, but now it appears that it won't actually give me any benefit.

Actually, for a 351 trivial (the trivial for radiant cut augments, IIRC), max success rate occurs at around 294 skill, assuming no masteries.

When they upped the max skill to 300, they effectively nerfed chanters, since JCM provides negligible benefit now.

It is possible though that they may up the trivials when trivials are adjusted, but that is just rampant speculation on my part, and I am not holding my breath that they will do so...

Aeght

[Taushar]

I had a few minutes of discussion with Absor after it was over. I made my case about those of us who are not in uber guild can not have the same access to the items to skill up. I am level 70 with 250 or so AAs but I still don't have elemental access. I told him that things like Shadowscream I am willing to farm, eventhough it takes forever. However, the exception of few items, all the ways to skill up from 250 to 300 requires keys, flags or raids. Even the rare items he mentioned as better chance to skill up from require very high level to get.

I had few other example for him. At the end he said that wait until the new tradeskill modification comes out before talking to him again. However, I think he kind of got that some of us want a skill up path that may be coincide with uber item creation. Not all of us want to make high end items as the ultimate goal of doing tradeskills. We want achievements but if they restrict us to making high end item as the only way to skill up, that leaves many of us behind.

I will be talking to him at the next fan faire after everything settles down and see if things will be better or not.

Taushar

[callius]

hmm what skill up patchs require flags/keys that are unreasonable to someone of your lvl not in a raiding guild ?

[YurienStonebow]

well i'm not in an uber raiding guild. my guild doens't ahve anyone flagged for pop other then maybe hedge maze completed. we can't farm pop for ingredients so a lot of recipes are out for us

[Bobaten]

Actually, for a 351 trivial (the trivial for radiant cut augments, IIRC), max success rate occurs at around 294 skill, assuming no masteries.

When they upped the max skill to 300, they effectively nerfed chanters, since JCM provides negligible benefit now.

It is possible though that they may up the trivials when trivials are adjusted, but that is just rampant speculation on my part, and I am not holding my breath that they will do so...

Aeght

Well, the problem is, while the formula shows a max success at 294 skill, the actual success rate is much lower. It seems that the augment combines are examples of combines that have success rate capped below 95%. It now looks like the mastery aa's won't help to raise the success rate on recipies with these success caps. If the grandmaster armor combines also have these success caps, then the same applies there.

[Aeght]

Well, the problem is, while the formula shows a max success at 294 skill, the actual success rate is much lower. It seems that the augment combines are examples of combines that have success rate capped below 95%. It now looks like the mastery aa's won't help to raise the success rate on recipies with these success caps. If the grandmaster armor combines also have these success caps, then the same applies there.

That would make JCM even less useful, rather than more useful...

I believe GM armour trivials are high enough to make tailoring and smithing masteries useful (unless the cap makes them otherwise).

If this is the case, though, why are we being charged 18 aa's for little to no benefit? Shouldn't the costs be at least lowered?

Frankly, unless there are either sufficiently higher JC trivials set or new recipies which are better than OoW augs (which would take quite a bit to accomplish and may well be impractical), enchanters should be compensated somehow...

Aeght