# 23 Jan, 2020 23:15
With no new image since a week and 15 more likes since 24 hours, my index still always the same ( 98,91 )
most of these likes are for a two years old image ….
Is there a reason for that ?
Thank you for your answer.
# 24 Jan, 2020 11:03
Given the recurring posts on the AB index, I became curious to know how that number is computed. I'm writing this in the hope that it helps those who do not speak python. The source code tells me that the index I is such that (Salvatore correct me if I'm wrong):|
I = log(N+1) x Average(likes >= A)
A being the average of the likes of all images, and N the number of images with numbers of likes greater or equal to A.
Taking it lightly, here is an example of how it can behave. Say you have three images in total, with 10, 15 and 20 likes. The average number of likes is A=(10+15+20)/3=15, and N=2 (two images with indices >= A = 15). The average index of those two images (the two best) is 17.5, so I = log(2+1) x 17.5 = 8.34.
Now you post a new image that everybody likes very quickly, so the other images don't get new likes during my experiment. The new image starts at 0 likes, so A = (0+10+15+20)/4=11.25. The number of images with indices >= A does not change, so I = 8.34 still.
For each new like on that new image, we have:
likes | Global average | N | Average of N best | I
1 | (1+10+15+20)/4=11.5 | 2 | 17.5 | 8.34
… | …. | … | … | …
14 | (14+10+15+20)/4=14.75 | 2 | 17.5 | 8.34
15 | (15+10+15+20)/4=15 | 3 | 16.67 | 10.03
So far so good. Likes do not have any impact until they affect the global average enough to change the number of best images. The average of the best ones actually decreased at the last step, but that was compensated by the increase of N. Let's continue:
16 | (16+10+15+20)/4=15.25 | 2 | 18 | 8.59
I got more likes and my index decreased! Let me complain! Just kidding The global average A increased so that he number of best images is back to 2, with an averaged number of likes up to 18, but that is weighted down by the log (2+1) term. Now just wait for more likes …
… | …. | … | … | …
23 | (23+10+15+20)/4=17 | 2 | 21.5 | 10.26
It got back up, what a relief! Until 35 likes on that new image, N will be 2. But wait …
… | …. | … | … | …
35 | (35+10+15+20)/4=20 | 2 | 27.5 | 13.12
36 | (36+10+15+20)/4=20.25 | 1 | 36 | 10.84
Back down again! Let me complain! Just kidding The new image is liked so much that it pulls the global average high enough so that it became the only one above average!
I'm stopping here. A few more likes and the index will go back up to where it was again.
While playing I could see my index evolve exactly as predicted by the above formula, so no surprises here. Note that the update is not instantaneous and that I've seen it occasionally go down to zero or some weird value, but these glitches don't last and I suppose they are due to client/server cache behaviors.
The threshold effect of taking the average of the images above average creates fluctuations, but these are smoothed out in the long run. They are also amplified in the example above because of the small number of images considered.
Anyway that's the way it works (unless I've missed something). Whether it is meaningful or not is another matter, I'm not going into that.
Frédéric [should be doing something else ]
# 24 Jan, 2020 11:05
|I don't have a quick answer for that, but since I get so many questions (really, daily) about the index, I'm going to add more plots to user profiles to see daily likes and index changes. Hopefully this will add some transparency behind what is really just a mathematical formula.|
# 24 Jan, 2020 11:08
I only had a quick read and the table is not formatted, but that looks spot on! Can you try to fix the tables? Use the [BBCODE] button in the top-right of the form text area if necessary.
I will try to add this to the FAQ.
# 24 Jan, 2020 11:38
I tried to format te tables in either HTML of BBCODE, but could not get it to work. Replaced with standard text for now. I tried to finesse this but failed.
That keeps happening to me
|You have no new notifications.|