101 replies to this topic

[RJEM]

You need a lot more tests if you want to prove something...

I thought I'd shown this to be false by using some real statistics rather than perception :roll:

Real data, anyway. An attack population of 333 doesn't provide much in the way of statistics.

Although, I do agree with ya about deflect. If you DID run a lot more tests of this, I'd be willing to bet that the activation percentage would go up. to the range of 43.75%, like the skill description says it should; not "ZOMG, deflect is bugged" variety.

I both agree and disagree slightly here - whilst 333 doesn't nail down the fixed percentage very well, it IS enough to rule out the much higher percentages being claimed by some (to a very high probability) - so it is statistically significant to a degree

I would also bet that the rate would tend to 43.75% in the longer run - I was demonstrating that this is the case and mentioned that my rate was slightly low, but within expected/allowed variation around that value - which 50% would not be

Anyway, was a bit of overkill, but it shows the dangers of just looking at extremely poor or good runs for your information.

[Prezze]

Hmmm, as a part of the guild Gutbusters, I should help another fellow buster .

Seems nothing wrong with the statistics on deflect in my opinion.


On a sidenote for your idea on the find item buff => you have to count every drop, common to legendary.

Also the droprates of regular and legendaries does take a roll in here.

Personal experience with scavenging

fi1000 > fi800

[Uncle Beg]

So, I was extremely bored this evening and decided to do a little experiment with Deflect and find out whether the horror tales of neverending bounties and frustrated GvGers had any substance to them besides being just bad luck.

To do this, I equipped myself inadequately and went out to do 3 GvGs against large guilds (to save me some time!). I chose For Fracks Sake, The Eternal Warriors and angellife thanks mainly to their large number of targets in my range and also because they are conveniently at the top of the ladder Many thanks to them for their assistance in this.

Every target was buffed with Deflect 175 prior to me beginning the conflict and I wasn't concerned about the actual result of each fight - just the rate of activation of deflect. As we know from the skill description, Deflect 175 should result in a failed attack 43.75% of the time. Keep that number in the front of your minds.

On to my findings:

1st Conflict: FFS

100 successful attacks required 158 attempts on my part (incidentally I won 26 times )
Deflect Activation Rate: 36.71%

2nd Conflict: TEW

50 successful attacks required 94 attempts on my part (10 wins here, go me!)
Deflect Activation Rate: 46.81%

3rd Conflict: angellife

50 successful attacks required just 81 attempts (with 6 wins )
Deflect Activation Rate: 38.27%

_____________________________________
Overall Totals

333 attempted attacks
133 deflected attacks

Deflect Activation Rate: 39.93%

This is a reasonable number - but is it perhaps lower than it should be?

The p-value for 200 attacks from 333 attempts (assuming a null hypothesis with a 43.75% probability of failure for each attack) is 0.089. Although low, the null hypothesis would *not* be rejected at this level in most scientific fields. Essentially, a 40% activation rate over 333 attempts is within acceptable random variation about the hypothesised rate.

Just so all of you bounty hunters feel happier, I did have a look through my raw data and found the following pair of sequences. I assumed I would need 10 successful attacks to complete the bounty, and I can feel your pain. a = Attack, d = Deflect.

1. ddaddaddaaddadddddddaddddaaaa = 29 hits to get 10 kills
2. ddddadddddddaddaaddaddaddaaaaa = 30 hits to get 10 kills

However, in the long run I actually got quite lucky and had my share of long winning streaks too. This was the best I could find:

1. aaadaaaaddaaaaaaaadadddaadaaaaddaaaaa = 37 hits with only 10 deflected

My advice: Be positive. If you're in a bit of a mire just keep dreaming of the day you breeze through deflect like it wasn't there. It will happen - eventually!

Next up, I'll be taking a look at whether FI 800 really is better than FI 1000 when scavenging. I just need to find some (lots of) gold before I start

Very nice post RJEM!

If you have time, could you try Last Ditch and Force Shield.

Thanks.

[aa0007]

hmm .. if you are going to test scavenging, it would be fine to do it with just the doubler monster, 2500 gold a shot.

If you're going for 500 scavenges with FI 800, and 500 with FI 1k, that'll just be 1.25 mil gold per, so 2.5m gold for both, meaning around 16-17 fsp worth of gold. Not much

[RJEM]

hmm .. if you are going to test scavenging, it would be fine to do it with just the doubler monster, 2500 gold a shot.

If you're going for 500 scavenges with FI 800, and 500 with FI 1k, that'll just be 1.25 mil gold per, so 2.5m gold for both, meaning around 16-17 fsp worth of gold. Not much

I think the problem I'm going to encounter scavenging-wise is that you really want a standard number of drops from the same creature, and then vary the FI level to work out how often it kicks in above some base value.

500 scavenges on some of the rarer items would only get 1 or 2 drops even with FI 1050 running, so the chance of some really big variability between runs is huge (as we see every scavenging event). If, on the other hand, you obtain 200 of those drops then you'll be averaging the rate over '200/drop rate' combats, which will be a much, much bigger sample size but also correspondingly more expensive

I do like the idea of using the Cave Skeleton and the greater doubler potion as a test case though - to work out a methodology. 1.25% drop rate means about 16000 combats for 200 drops without FI active - or 40 million gold. Obviously there is all of the other trash in there to see as well, so maybe a factor of 5 to account for non-study creatures? I'd say the whole thing could cost about 400 million gold overall if done properly.

[Zukira]

Now try with 165 or 170, I swear it activates more!
(not even kidding, I find I have much poorer luck with the supposed lower %, but that must just be me)

I know what you mean - I'm more apt to ask someone who has a 165 for a deflect in a gvg situation than someone with a 175.

[RJEM]

Well done, RJ.

However, I think it's useful to point out that while in the long run, things may wash out, certain players in certain instances have been shown to have horrific luck.

Your numbers are effective in long range long term analysis, but I'd be curious what would happen if, for example, I ran numbers. I know inventors and bounty hunters who have either great (Evilaltor) or horrible (StickIsGod) inventing luck, and bounty hunters who have awesome times with deflect (me, on a good day) and insanely bad luck (me, on a bad day) with deflect.

I guess what I'm trying to say is that if you look at the RNG for Fallen Sword itself, it's not really random. It shows a distressing propensity to remember past success or failure. If you can dig up some of avvakum's old posts, one or two of them display photographic representations of the RNG, and it is most definitely NOT strictly random.

Yep - I think I mentioned that overleaf when talking about streakiness. If you happen to be in a bad patch there is a chance that it will continue. I think StickIsGod is one of the more extreme examples with his inventing, but my luck with the same thing has been much better than warranted.

I would guess that a) over the course of your PvP career it's close to 43% and you've had some pretty bad days mixed in there. Those representations avvakum posted were a good way of showing it graphically. I just wanted to show that in the long run (including every player and every attack) there is nothing wrong with deflect and it is activating as intended.

At the end of the day, you have better odds of getting 10 hits without deflect triggering than you do of flipping ten heads in a row with a coin. That's as it should be Any streak can come to an end at any time - and if Stick tried 75 more Ward part Ds he might regress towards the 50% success you would expect, and Evilaltor might go the other way. Luck is a fickle friend.

[aa0007]

hmm .. if you are going to test scavenging, it would be fine to do it with just the doubler monster, 2500 gold a shot.

If you're going for 500 scavenges with FI 800, and 500 with FI 1k, that'll just be 1.25 mil gold per, so 2.5m gold for both, meaning around 16-17 fsp worth of gold. Not much

I think the problem I'm going to encounter scavenging-wise is that you really want a standard number of drops from the same creature, and then vary the FI level to work out how often it kicks in above some base value.

500 scavenges on some of the rarer items would only get 1 or 2 drops even with FI 1050 running, so the chance of some really big variability between runs is huge (as we see every scavenging event). If, on the other hand, you obtain 200 of those drops then you'll be averaging the rate over '200/drop rate' combats, which will be a much, much bigger sample size but also correspondingly more expensive

I do like the idea of using the Cave Skeleton and the greater doubler potion as a test case though - to work out a methodology. 1.25% drop rate means about 16000 combats for 200 drops without FI active - or 40 million gold. Obviously there is all of the other trash in there to see as well, so maybe a factor of 5 to account for non-study creatures? I'd say the whole thing could cost about 400 million gold overall if done properly.

Instead of trying to get 200 drops, because thats gonna be a pain in the ass to do with no find item, why not just have a larger sample size, say 1000 or 1500 scavenges? That should give you a somewhat "even" value, and then just run the same number of scavenges on fi 800, and fi 1000.

[dowuones]

Ive just finished a test for FI

First Creature with 2 drops, 3.75% each (according to UFSG)

no FI: 1000 combats, 70 drops (7%)
FI250: 1000 combats, 254 drops (25.4%)
FI800: 1000 combats, 672 drops (67.2%)
FI1000: 1000 combats, 830 drops (83.0%)

Second Creature with 4 drops, 2.5% each (accordind to UFSG)

no FI: 1000 combats, 102 drops (10.2%)
FI250: 1000 combats, 359 drops (35.9%)
FI800: 1000 combats, 902 drops (90.2%)
FI1000: test stopped after 500 combats and 500 drops (it was a bit annoying )

Result: F1000 is better than FI800 (if there was any doubt)

My supposition (at least i think that is how FI works) is that if an item has a base of 2% chance drop rate, FI doesnt multiply that percentage, but adds chances to the base percentage, but now my bad english becomes incomprehensible, so an example might be better:

if drop rate is 2%,
with FI250 you have 2% (base) + base*2.5 (FI) = 7%
with FI800 you have 2% (base) + base*8.0 (FI) = 18%
with FI1000 you have 2% (base) + base*10 (FI) = 22%

Again this is just a supposition, but during all the test ive always seen numbers close to that.

(P.S. and now, after 3,689 drops destroyed i wont scavenge anymore for a couple of years :wink: )

[RJEM]

Ive just finished a test for FI

First Creature with 2 drops, 3.75% each (according to UFSG)

no FI: 1000 combats, 70 drops (7%)
FI250: 1000 combats, 254 drops (25.4%)
FI800: 1000 combats, 672 drops (67.2%)
FI1000: 1000 combats, 830 drops (83.0%)

Second Creature with 4 drops, 2.5% each (accordind to UFSG)

no FI: 1000 combats, 102 drops (10.2%)
FI250: 1000 combats, 359 drops (35.9%)
FI800: 1000 combats, 902 drops (90.2%)
FI1000: test stopped after 500 combats and 500 drops (it was a bit annoying )

Result: F1000 is better than FI800 (if there was any doubt)

My supposition (at least i think that is how FI works) is that if an item has a base of 2% chance drop rate, FI doesnt multiply that percentage, but adds chances to the base percentage, but now my bad english becomes incomprehensible, so an example might be better:

if drop rate is 2%,
with FI250 you have 2% (base) + base*2.5 (FI) = 7%
with FI800 you have 2% (base) + base*8.0 (FI) = 18%
with FI1000 you have 2% (base) + base*10 (FI) = 22%

Again this is just a supposition, but during all the test ive always seen numbers close to that.

(P.S. and now, after 3,689 drops destroyed i wont scavenge anymore for a couple of years :wink: )

Nice backpack abuse, and good gold sink

I'll have to think of something else to do for part 2 of Mythbusters now

Edit: By the way - that does mean the description for FI is completely wrong HCS. Doubling the chance of a drop makes sense according to the description - adding 10 times the base chance doesn't really as it's a 1000% increase rather than the expected 100%.

Looks like you guys have a decimal point in the wrong place? Ticket submitted anyway.

[SirAdmiral]

How is that low? That's pretty high. The stamina it costs to cast the buff is well worth it for that many deflects.

[RJEM]

2. ddddadddddddaddaaddaddaddaaaaa = 30 hits to get 10 kills

dddddddadaddddaddaaaadaddada.

Seriously, this was a complete joke.

And it was level 155 deflect.

28 hits to get 10 kills. And that's assuming I didn't lose any, which I did, as the target was online, fully buffed, cloaked, and had 48 levels on me, not to mention two fully empowered relics.

It happens - it's pure chance Gravely and I know you understand that. Doesn't diminish the frustration, and I definitely wouldn't want to try a tough bounty with Deflect 200 active. Would be silly.

[RJEM]

How is that low? That's pretty high. The stamina it costs to cast the buff is well worth it for that many deflects.

How is what low?

If you mean my original post, 39.9% is a bit low compared to the expectation value of 43.75%. It's actually a buff which activates a lot for the stamina cast - hence the HUGE frustration it can cause.

[RJEM]

Apologies - it's a computer implementation of pure chance which isn't brilliant and tends to clump!

That should cover it.