Hi all,
I've been wondering with all of the recent messages to the list about AWL if there wasn't a simple way of improving the system somewhat by including some information about the variance or std. deviation of the scores. Either can be calculated from the sum of squares of the scores and the average.
Basically what I'm thinking is that one could reduce the effect of AWL if the variance is high w.r.t the average, and vice-versa if the average score is very solid.
i.e. if someone routinely sends you messages in the range from -4 to -5, but then sends you a +8, do AWL agressively like today.
However, if someone sends you a single -5 message followed by a a +9 do no correction at all since the variance is infinite.
After the second message the variance would be smaller, but still much larger than the average score, so still keep AWL down.
Does this make any sense at all? Has this previously been considered and thrown out?
Thanks,
Balam
I've been wondering with all of the recent messages to the list about AWL if there wasn't a simple way of improving the system somewhat by including some information about the variance or std. deviation of the scores. Either can be calculated from the sum of squares of the scores and the average.
Basically what I'm thinking is that one could reduce the effect of AWL if the variance is high w.r.t the average, and vice-versa if the average score is very solid.
i.e. if someone routinely sends you messages in the range from -4 to -5, but then sends you a +8, do AWL agressively like today.
However, if someone sends you a single -5 message followed by a a +9 do no correction at all since the variance is infinite.
After the second message the variance would be smaller, but still much larger than the average score, so still keep AWL down.
Does this make any sense at all? Has this previously been considered and thrown out?
Thanks,
Balam