1.51
#...
·
3
likes
|
|---|
|
Imagine you logged in one morning and had 229 new likes! You look at your AB Index and it has actually dropped by 0.27 points! You would probably send a message to Salvatore asking him to fix it. BUT it might not be broken.... The most important point to understand about the AB Index is that all likes are not equal. Only the images with at least the average number of likes will contribute directly to your AB Index. That's right, if an image doesn't get to your average number of likes it doesn't contribute to the index, but it can actually lower your index by a small amount by increasing your average number of likes. I have written a spreadsheet (1, 2) and tested the scenario I described above, 229 likes can actually decrease your index. Here is how that happens and an explanation of why this is NOT a bad thing. In this scenario all 229 likes were for existing images that are below your average number of likes per image; therefore they do not contribute to your index value directly. Worse yet they increased the average number of likes so some images that used to be above average are now below average and no longer contribute to the AB Index. As a result your index goes down slightly even though you got 229 new likes! Nothing is broken, that's the way it is supposed to work. This may seem wrong, but this actually encourages posting failures, what went wrong images, and experiments that failed. Often these types of images get few likes so you might not post them; however the current index has a much lower penalty than an index based on a simple average. To prove that the present index is better than a simple average I ran a scenario in which I added 20 more posts to my "virtual Gallery" of unloved images such as photos of failed attempts at weak DSO's or images of M40, that averaged just 15 likes each. None of these images reached my average number of likes so they had a minimal negative effect on my AB Index, changing it from 61.40 to 61.19, a change of -0.34%. Using a hypothetical index based on a simple average of all images the index changed from 27.74 to 25.6, a change of -7.7%. Whether you like the AB Index or not it is probably here to stay. It may not be perfect, but it rewards good images and has an insignificant penalty for risk taking. Overall it may be one of Salvatore's best ideas. 1) I used my own data from 25 Sep, 2020. AB Index 61.4, 99Images, average likes 27.7 2) My spreadsheet has been verified correct by Salvatore. 3) I have decided not to release my spreadsheet. The index given on the AB site is not a real time index so it always trails the real time index reported by my spreadsheet. I am afraid that releasing my spreadsheet would risk a flood of "My index isn't right" complaints. |
11.07
#... |
|---|
|
Thanks David. The index behavior question pops up frequently so it’s good to have a concise explanation. At one point I tried to think of a formula change that would maintain all of the positives of the current system but removes that small score drop when a new image is uploaded. That’s what always gets people riled. But I could not think of a better way. |
|
1.51
#... |
|---|
| Thanks Kevin! Just trying to make AB a better neighborhood. |
10.46
#... |
|---|
| David, thanks for the clean explanation of the index behavior. Best regards, Vitali |
2.11
#...
·
 |
|---|
|
Thanks David, I too also , as a curiosity derived from my background of physicist, have been able to replicate the astrobin index calculation within an Excel spread-sheet. For example I have been able to replicate the exact value as it appears if considering my gallery consisting (until now) with 21 images. Thanks to the github open-source here "https://it.welcome.astrobin.com/features/image-index" the formula used in astrobin is very clear. Even the log_10 function on the effective number of images +1, the latter triggered by the present average thresold, then multiplied to the average of the sum of all the images above that thresold. Best, Davide. |
