Reposting my conclusions:
Quote:
Originally Posted by bcbrown
[You must be logged in to view images. Log in or Register.]
There's a 1 in 6 chance of a stun happening pre-slow.
If there's a slow, the expected damage is 171.
There's a 3/17 chance the expected damage is either 320 or 448.
This means the worst case will happen around .1725 * 3/17 or 3% of the time.
|
Comparing short-time intervals, which I happen to think is mostly irrelevant (as pointed out by Jimjam), but has been requested:
Quote:
Originally Posted by bcbrown
[You must be logged in to view images. Log in or Register.]
Expected damage saved pre-slow by FSI overall:
82.75% of the time: none
17.25% of the time: 133
Total:
0 * .8275 + 133 * .1725 = 22.9425
|
This analysis can be improved by incorporating a resist rate for Turgur's, and by replacing the use of DPS with a damage distribution that incorporates the possibility of a damage spike far exceeding average DPS.
If, for example, there's a 10% chance of actual damage being twice DPS over the relevant time period, I think there would be a 0.3% chance of 600+ damage.
To build a damage distribution from a log file, first filter out all lines except where you take damage. Then, for each line, remove everything except the timestamp and the damage taken. To bin the data, first choose a bin size. Reasonable possibilities could be 10 seconds or one minute. Transform the timestamp using the bin size, e.g. at ten seconds, turn 10:53:43 into 10:53:4. Group the data by bin, and compute the total damage by summing all the entries.
Now you have your observed damage distribution. To construct a modeled distribution, first choose the model; normal seems likely to fit. Then use a software package to calculate the best-fit for the model and confirm it's a good fit with a couple statistical tests. I'd have to further research this part, but
https://www.datanovia.com/en/lessons...ribution-in-r/ seems like a good tutorial in R.