Thursday, August 20, 2009

Arena Fizzles Part II (K.I. - Fix This!)

I kept track of fizzles the other day in arena. My death noob fizzled on Vampire (an 85% spell) 4 times out of 9 casts (65% actual performance). He also tried to heal himself with a 90% treasure life spell which fizzled no less than 3 times IN A ROW before it finally cast. (1 out of 4 casts is 25% not 90%)

On my ice wizard I fizzled twice on Frost Giant (an 80% spell) out of about 6 casts which would seem to be passable IF my clothes didn't give me an additional 6% accuracy. So that's a 66% performance for an adjusted 86% spell.

I did however cast a 70% Triton 3 times without a single fizzle (100% actual performance).

When you're casting death spells and you're holding your breath because you're expecting them to fizzle... there's something seriously wrong.

I'm asking KingsIsle to fix the spell percent performance. At this point the higher the percent spell the more likely it is to fizzle in arena. Either that or actually adjust the numbers on the cards so that they reflect reality.



Lost said...

Though I agree with the fact that the fizzles have drastically gotten out of hand, I have to say a couple things. First off, yes, your percentages might be correct, when comparing them to each other, in one battle, but the percentages on the card are not supposed to reflect the over all accuracy, but the accuracy each card per turn has.

Like with Storm. 70% or 7 out of 10. Take a ten sided die and roll it. Basically, if that die lands on 1-7 it's a go, if it rolls an 8,9, or 10 it's a no go. Or a fizzle. The percentages are based on indivual turns, not over all.

Again, like I said before, the fizzles have drastically gotten out of hand, and I do think something needs to be done. But for all we know, maybe something had been done. Just not in the way we like?

Sierra Starsong said...

How many fizzles are you getting out of larger samples, say, fifty casts or a hundred? Playing both Life and Storm, I seem to have more fizzles against street mobs than from PVP.

Valerian Nightbringer said...

I agree with the analogy of the ten sided die and that percentages are turn based. However you will find that as you continue to roll the die and the number of rolls increase toward infinity the results should get closer and closer to the actual percentage listed IF the percentage is truly 70% for example. The disparity that i've noted is former percentage performance against recent percentage performance which if you've been in arena you probably notice that it's nearly flipflopped, the higher percentage the spells are performing only equal to the medium percentage spells.

One thing that programmers often don't realize is the Psudo Random Number Generators are not truly random. And there are few PRNG that actually work well. In the end all are simulated and therefore while they can come close to true randomness, without integration into the physical world you will never be able to achieve truly random results.

To address this I'd recommend a connection to to pull noise data to seed the PRNG. When a PRNG does not get seeded properly it's results are eventually predictable and therefore what would normally be statistically sound assumptions (a go or no-go result will actually show x% of the time ) Like the software in slot machines the PRNG can be biased toward specific results based on predictable seeding.

Valerian Nightbringer said...

Also of note is that a standard PRNGs produce results within a number range via truncation or rounding. Truncation and rounding "adjusts" a larger number down to a smaller number taking a more accurate supposedly random number and thunking it lower or higher to fit the bucket that has been defined... for example if the PRNG produces .8500172 as a result and the programmer wants a number between 1 and 8 they call the PRNG with a parameter that tells it to truncate the results to their desired range. If truncate is used the eventual result is 8. If round is used, the number will be returned as 9. Now this means that the randomness of the PRNG is compromised where these thresholds are crossed and the expected pre-assumed percentages estimated are no longer correct.

Now with a fizzle, no fizzle result like what was proposed about the die casting method... a larger percentage number has to be boiled down to a number between 0, and 1. The die method is often the means to achieve this since it allows for an approxiate percentage result to be boiled down to a binary result. Now if a PRNG is biased toward low or high numbers you may say any number above 2 is a GO and 2 and below are NO GO results. What happens if the PRGN likes to produce numbers between 1 and 2 more often than 9 and 10? It means that the anticipated percent is not accurate. If you combine that with a truncation or rounding bias you can be looking at a fizzle percentage = to 30% instead of 20%. The combine that with the concept of truncation/rounding whereby the upper and lower bounds are set the PRNG now cannot produce results below the lower bound this means that the probability of a one being produced is less than the probability of a 2 because a 2 can be produced by a larger number from the PRNG that is slightly lower then 2, equal to 2 or a little greater than 2, (three chances) while a 1 can be produced only if the result is equal to 1 or slightly greater than 1 (two chances). These types of biases are what what I believe are impacting the game's results if not an acutal on purpose manipulation to handicap schools with high percentage spells.

Ignoring truncation issues for a moment, if we are talking having the highest quality random number generator there are specific algorithms for encryption that have been noodled over by the best minds in the business. Yarrow is a good one, Mersenne Twister is touted a lot... however these are still not truly random unless seeded by actual physical world data. If you want to get really fancy you can use which provides a physical data PRNG seed feeder which allows random number generators to actually be random when seeded by feed files.

What can be done to fix this?

The percentage calculators altorithms need to be run through a unit test to excercise them significantly and the results dumped to statistical analysis to ensure that over hundreds or thousands of casts these do in fact perform at the expected level.

But as noted before, when minor strikes and shields (100% spells) occasionally fizzle without the aid of a minus accuracy there is a problem and the fizzling of high probability spells becomes drammatically noticeable after a game update it makes one wonder why has this suddenly changed?

Also of note was the failure of beguile, once cast, to make the wizard cast on their own team, which failed twice in such a way that while the spell took, the victim was still able to wild bolt, and in another instance, cast storm lord on the on the opposing team.

In my opinion all of these anomalies starting from the date of the last program update point toward bugs or biases introduced by recent programming changes.

Devin ,Jacob and dugan gold grove said...

Hey Valerian you want to know what my blog is going to be about? Well here it is! Hi this is the master of balance Dugan GoldGrove along with my two twin brothers Jacob goldGrove and the new addition to the family Devin!
We will make vidios about all our adventures in wizard101. We will give game cheats and all that good stuff. Oh and one rule NO BAD POSTS WITH CURSES IN IT. We will be happy to be friends with people in our spare time. Good luck with the game. :)

Lost said...

I agree with you there, if you are correct in saying they do use that method. I was just under the impression that the game mimic'd rolling a die, with out any other factors involved. I think I understand what you are saying, and though I don't PVP I have been noticing a lot of my life spells fizzling, when my storm spells are hitting fairly commonly. However, my fire spells are still back to back fizzle, then hit, fizzle, hit, etc. This is all against street mobs and bosses though. I haven't noticed any 100%ers fizzling, but since you're in the arena, things might be slightly different.

Weren't they not too long ago supposed to be fixing the bug in PVP? Maybe this is also one way to fix it in their eyes? Make everything less predictable, and giving a some what equal advantage against other teams? Again, I really don't know, as I'm not into the PVP tournies much at all.

Dugan GoldGrove said...

Hmm Valerian some thing tells me your an adult. At the first time I visited here I thought you were a kid now that I read what you said you're an adult I think I'm not sure.
P.S. I did some Editing.
From Dugan.

Cheats and Crackers said...

I've noticed this more and more as well. In the arena when I duel, I have a myth on my team, because earthquake and double hits are needed so much. He makes earthquake treasure cards because when we really need it to hit, we don't want it to fizzle. Yet with around eight or nine accuracy boost from his clothes, it still fizzles. With the boosts it probably has a 4% chance of fizzling, yet what has been happening is it changes it to like a 60% chance of fizzling. Kingsisle really does need to fix this ASAP.

Valerian Nightbringer said...

Last night, using my old zeke cloths, (+9% accuracy) I fizzled putting on a storm/fire shield (supposedly 100%) which after getting storm lord cast on us, drammatically changed the outcome of our battle.

Valerian Nightbringer said...

Yes I am an adult. My regular work is basically being a hacker for a company, testing security systems. When I do programming I usually work on security stuff like encryption, artificial intelligence, or predictive modeling where my programs predict real world things like what the probability is that a person will get sick and how much money they will spend next year visiting the doctor.