As trials continued, so did others, until we ended with a solid six points outside of the interval, a likelihood too small to count if True Hit stayed the same. Suffice to say, the very first value we collected a significant number of points for (75) quickly fell outside of the confidence interval for an unweighted 2-RN model (A+B)/2. This means that when testing 99 different values, we should see possibly 1, maybe 2 “actual” values outside the confidence interval for any given model. I used a 99% confidence interval, meaning that the system should fall within the bounds 99% of the time. These are normally based around the normal distribution curve, and since this is a binomial system (i.e., this is a system with two outcomes per trial, hit or miss), the same applies here. Some background as to why we know it’s dead: in statistics, we have something called confidence intervals (CIs), which are the bounds between which we can expect the system to actually be based on what we measure. This led to 0.03% chance at Hit = 1, with a solution set of (0,0), (0,1), and (1,0) out of 10000 possibilities. That version used two random numbers between 0 and 99, “A” and “B”, averaged them, and compared them to the displayed value to get a result. True Hit as we know it is, to 99.999999% confidence, dead. Remember when I gave the Demystifying the Math presentation on True Hit and called it the bastard child of a logistic curve and Fomortiis? The new system used in Fates is what you would get if you bred Fomortiis with Anankos and paired off the spawn with a logistic curve. 23000 points later, and hey, we’ve got something semi-conclusive.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |