Random attacks or moves

[STM1993]

Gravedig because I have some actual numbers from testing, done by doing 100 pairs of tests for each number of arests stacked(the maximum is 5 in 2.0a exe).

Method for getting the numbers (Click to View)

I used 2 Deeps to cast D>A at each other at the same time, with a modified deep_ball, 100 times:

DC-Code: <frame> 0 flying pic: 0 state: 3000 wait: 1 next: 0 dvx: 9 dvy: 0 centerx: 40 centery: 41 hit_a: 0 hit_d: 0 hit_j: 0 itr: kind: 8 x: 22 y: 27 w: 55 h: 27 dvx: 1 itr_end: itr: kind: 8 x: 22 y: 27 w: 55 h: 27 dvx: 2 itr_end: itr: kind: 8 x: 22 y: 27 w: 55 h: 27 dvx: 3 itr_end: itr: kind: 8 x: 22 y: 27 w: 55 h: 27 dvx: 4 itr_end: itr: kind: 8 x: 22 y: 27 w: 55 h: 27 dvx: 5 itr_end: <frame_end> <frame> 1 one pic: 7 state: 3001 wait: 5 next: 1000 dvx: 0 dvy: 0 centerx: 39 centery: 41 hit_a: 0 hit_d: 0 hit_j: 0 <frame_end>

MS Excel Sheet containing my results, using 200 pairs of results.


Ordered from top to bottom:
2 arests: 57% / 43% (so is it really 50-50 or 60-40? I don't know, but Nya says its 50%)
3 arests: 28% / 26.5% / 45.5%
4 arests: 11% / 10.5% / 28.5% / 50%
5 arests: 9% / 6.5% / 9.5% / 23% / 52%

As you can see from the pattern, it seems that LF2 is heavily biased towards the bottomost arest, followed by the 2nd lowest arest.
I can't really tell the odds of the remaining arests since I can't go beyond 5 arests in the 2.0a exe (in the old 1.9 exe, you could go up to 10 itrs), but the 2nd arest from the top seems to have consistently lower chances of happening than the 1st arest or the 3rd arest, while 1st & 3rd appear identical.

----

Also, as a bonus, a bit of info on stacking arest & vrest:

arest+vrest+vrest = both arest & 1st vrest hit, but only the end effect of the 1st vrest is visibly shown. The hitspark seems to be bugged however.
vrest+arest+vrest = only the arest and the 2nd vrest will hit.

This demonstrates two things:
a) arest is always checked first BEFORE vrest, and then it goes down from there.
b) Any vrest below the arest is then subsequently checked from top to bottom. Take the first one that can hit, ignore the others.

It should also be noted that you can use arest+vrest to consistently cause an injury bounce the same way that using Rudolf's D>A or Bat's D^J can do so.

Hope that helps.

[rewlf2]

As i recall some stuff in statistics, the error for a random sample is equal to 100/Sqrt(number of samples).

What this means is for 200 samples that you tested in the arest randomness, the random error is equal to 100/Sqrt(200), roughly equal to 7.07%

By taking the average data in Ttl5 in 5itr, Ttl4 in 4itr, Ttl3 in 3itr and Ttl2 in 2itr, the percentage for bottommost itr to take effect is 47.6+-7.07%, that is (40.7-52.6%)

I am inclined to what Nyamaiku said, the arest trigger couldn't be put into a ridiculous percentage like 45%. It is reasonable to think at followed:

2itr:
1st 50%
2nd 50%

3itr:
1st 25%
2nd 25%
3rd 50%

4itr:
1st 12.5%
2nd 12.5%
3rd 25%
4th 50%

5itr:
1st 8.333%
2nd 8.333%
3rd 8.333%
4th 25%
5th 50%

The mess i talked in the last paragraphs was just to show the statistic result fall into the percentages i proposed.

[Ramond]

My gut feeling says it will keep halving the percentage of the first itr each time. Akin to the probability of a number being chosen in the scenario: itr 1, yes or no? (50% chance of either) If no, then itr 2, yes or no? (50% of either) etc. That would make itr1 have 50%, itr2 25%, itr3 12.5% etc.

Yeah, the scenario I described is just for explaining purpose, not to describe what actually happens. But I believe such a pattern to be more likely than the sudden jump from 1/8 to 1/12 at 5 itrs.

Perhaps you can test the same thing in 1.9 if 10 itrs are allowed?