|
climbjs
Jul 1, 2003, 10:24 PM
Post #1 of 23
(3661 views)
Shortcut
Registered: Mar 1, 2002
Posts: 379
|
Hello. I am just wondering how a posted photo gets put into the rotation of photos on the homepage. Thanks.
|
|
|
|
|
cloudbreak
Jul 1, 2003, 10:30 PM
Post #2 of 23
(3661 views)
Shortcut
Registered: Jul 15, 2002
Posts: 917
|
You have to know the "special" people........the ones who iron their white socks at 1:30 am..........but don't tell anyone. :wink:
|
|
|
|
|
rrrADAM
Jul 1, 2003, 10:30 PM
Post #3 of 23
(3661 views)
Shortcut
Registered: Dec 19, 1999
Posts: 17553
|
It has changed a few times, but I believe currently: Top rated New Photos with 5 or more votes from the first 3 pages are randomly put on the FP.
|
|
|
|
|
climbsomething
Jul 1, 2003, 10:37 PM
Post #4 of 23
(3661 views)
Shortcut
Registered: May 30, 2002
Posts: 8588
|
The photo has to have a high rating. I don't know the formula, but it seems to change all the time anyway. This doesn't mean the photo always has to be GOOD, mind you, it just needs to have enough high votes...
|
|
|
|
|
allan_thomson
Jul 17, 2003, 10:38 PM
Post #5 of 23
(3661 views)
Shortcut
Registered: Jul 12, 2003
Posts: 596
|
In reply to: You have to know the "special" people........the ones who iron their white socks at 1:30 am..........but don't tell anyone. :wink: Is that 1:30 GMT, or what? :wink: :wink:
|
|
|
|
|
coldclimb
Jul 17, 2003, 10:42 PM
Post #6 of 23
(3661 views)
Shortcut
Registered: Jan 14, 2002
Posts: 6909
|
Adam: Is it the first three pages? I thought it was just the first page. :?
|
|
|
|
|
tim
Jul 17, 2003, 11:38 PM
Post #7 of 23
(3661 views)
Shortcut
Registered: Apr 4, 2002
Posts: 4861
|
In reply to: The photo has to have a high rating. I don't know the formula, but it seems to change all the time anyway. The formula for choosing the FP photo is documented in excruciating detail in this thread. Also, some rather insightful comments about subjectivity, composition, and the pros and cons of various ranking systems were well worth reading. Short version:
In reply to: 1) a random number X between 1 and 45 is selected (number of days since submitted) 2) another random number Y between 1 and 30 is selected (number of records to fetch) 3) after the top-ranked Y number of pictures (see above for notes on ranking) from the past X days have been selected (and increasing the # of days if necessary to collect Y number of records), a random number Z between 1 and Y is selected. The photo that has landed in Z'th place according to the sorting algorithm, is shown on the front page for the next 60 seconds. Note that, inasmuch as Biff has finished off the RMS averaging code, the 'score' is not the raw mean (sum of votes, divided by the number of votes) but rather the RMS (root(sum-of(votes^2)/number-of-votes)) score after ignoring outliers. This tends to ignore outrageously out-of-kilter votes and more clearly express the overall preferences of the majority of voters, hence it is preferred. (Another way to say it is that both puffing and bombing are severely curtailed by this method of calculating a photo's merit.) HTH, --t edit: clarified the definition of RMS
|
|
|
|
|
jt512
Jul 17, 2003, 11:45 PM
Post #8 of 23
(3661 views)
Shortcut
Registered: Apr 12, 2001
Posts: 21904
|
In reply to: Note that, inasmuch as Biff has finished off the RMS averaging code, the 'score' is not the raw mean (sum of votes, divided by the number of votes) but rather the RMS (root(mean^2)) score after ignoring outliers. Huh? sqrt(X^2) = X. You haven't accomplished anything. You sure you got that formula right? -Jay
|
|
|
|
|
tarzan420
Jul 18, 2003, 12:52 AM
Post #9 of 23
(3661 views)
Shortcut
Registered: Jun 19, 2002
Posts: 678
|
Root Mean Square is: sqrt( (i[1]^2+i[2]^2+...+i[n]^2) / n)
|
|
|
|
|
jt512
Jul 18, 2003, 1:10 AM
Post #10 of 23
(3661 views)
Shortcut
Registered: Apr 12, 2001
Posts: 21904
|
In reply to: Root Mean Square is: sqrt( (i[1]^2+i[2]^2+...+i[n]^2) / n) That gives large numbers greater weight. Therefore, high scores (whether deserved or not) will be hard to vote down, but low scores (whether deserved or not) will be easy to vote up. Was that the intent? One way to eliminate undue influence of extreme scores, whether they be high or low, would be to compute the median instead of the mean. -Jay
|
|
|
|
|
tarzan420
Jul 18, 2003, 3:27 AM
Post #11 of 23
(3661 views)
Shortcut
Registered: Jun 19, 2002
Posts: 678
|
If you are counting all the votes, that is true, but I am under the impression that outliers are eliminated; if this is the case the RMS is not so easily influenced. I don't know where the RMS comes from originally, I only know it from physics class (RMS voltage on AC circuit, etc.). I also don't know what the intent was on incorporating it into the photo scores.
|
|
|
|
|
jt512
Jul 18, 2003, 6:46 PM
Post #12 of 23
(3661 views)
Shortcut
Registered: Apr 12, 2001
Posts: 21904
|
In reply to: If you are counting all the votes, that is true, but I am under the impression that outliers are eliminated; if this is the case the RMS is not so easily influenced. If outliers are eliminated, then using the RMS seems pointless. The median seems is the best measure of central tendency I can think of to eliminate undue influence from outliers. -Jay
|
|
|
|
|
biff
Jul 18, 2003, 7:28 PM
Post #13 of 23
(3661 views)
Shortcut
Registered: Nov 5, 2001
Posts: 851
|
We are not really using a root mean squred algorithm. The algorithm is more like a median filter / average filter combined .. I don't want to discuss the exact implementation because people might be able to use that information to figure out how to manipulate the rank. But be assured that a photo gets a very accurate (in my opinion) rank based on the votes that have been sdubmitted for it.
|
|
|
|
|
jt512
Jul 18, 2003, 7:56 PM
Post #14 of 23
(3661 views)
Shortcut
Registered: Apr 12, 2001
Posts: 21904
|
In reply to: We are not really using a root mean squred algorithm. The algorithm is more like a median filter / average filter combined .. I don't want to discuss the exact implementation because people might be able to use that information to figure out how to manipulate the rank. But be assured that a photo gets a very accurate (in my opinion) rank based on the votes that have been sdubmitted for it. I'd be more likely to believe it if you knew the difference between a rank and a rating. But seriously, I don't know if Tim mentioned it to you, but about a week ago, I gave lowish ratings to two highly rated photos, and the average rating went up. I don't know if you've fixed that, but if not, then your algorithm has a bug. -Jay
|
|
|
|
|
enigma
Jul 20, 2003, 5:38 AM
Post #15 of 23
(3661 views)
Shortcut
Registered: May 19, 2002
Posts: 2279
|
In reply to: In reply to: We are not really using a root mean squred algorithm. The algorithm is more like a median filter / average filter combined .. I don't want to discuss the exact implementation because people might be able to use that information to figure out how to manipulate the rank. But be assured that a photo gets a very accurate (in my opinion) rank based on the votes that have been sdubmitted for it. I'd be more likely to believe it if you knew the difference between a rank and a rating. But seriously, I don't know if Tim mentioned it to you, but about a week ago, I gave lowish ratings to two highly rated photos, and the average rating went up. I don't know if you've fixed that, but if not, then your algorithm has a bug. -Jay I think I know the secret, they drop the score if its the lowest one. So if the photo is a 9.50 and you give it a 8 and thats the lowest, your score gets dropped.
|
|
|
|
|
apollodorus
Jul 20, 2003, 5:42 AM
Post #16 of 23
(3661 views)
Shortcut
Registered: Feb 18, 2002
Posts: 2157
|
Put a great photo out there, and the rest will take care of itself.
|
|
|
|
|
thomasribiere
Jul 20, 2003, 11:34 AM
Post #17 of 23
(3661 views)
Shortcut
Registered: Aug 24, 2002
Posts: 9306
|
^right! The formula seems so complicated that I can only be very satisfied and greatful to all of you when I see one of my pics on the front page. So thank you all for appreciating some of my pictures! THANK YOU. :D
|
|
|
|
|
ambler
Jul 20, 2003, 1:40 PM
Post #18 of 23
(3661 views)
Shortcut
Registered: Jul 27, 2002
Posts: 1690
|
In reply to: The median seems is the best measure of central tendency I can think of to eliminate undue influence from outliers. Ho ho. It's simple but not "best" in the minimum-variance sense. I bet we can think of others. 8)
|
|
|
|
|
biff
Jul 20, 2003, 3:37 PM
Post #19 of 23
(3661 views)
Shortcut
Registered: Nov 5, 2001
Posts: 851
|
rank is a value to describe position in a list rating is a value used to to be able to compare things in a more general way. I used the word rank becuase that is what is used in the database, and that is the term used in the filter (i.e. sort by: Rank). I agree that rating would probably be a better term, but I think it is prety easy to figure out what we mean by rank, so I'm not changing it, but I will use the word rating for the rest of this post. As to why your vote actually increased the rating .. one of 3 reasons .. the first being the most probable, the last two being very unlikely. 1) The algorithm changes when there is more than 4or 5 votes on the photo your vote pushed it over the threshold and it recaclulated the rating based on the different algorithm. 2) The rating stored might have not been updated when we implemente the new algorithm a few months ago, so when you cast a vote the new rating was calculated based on the new algorithm. 3) There is a bug in my code of which I am not aware.
|
|
|
|
|
tim
Jul 20, 2003, 5:40 PM
Post #20 of 23
(3661 views)
Shortcut
Registered: Apr 4, 2002
Posts: 4861
|
In reply to: 3) There is a bug in my code of which I am not aware. Just in case this actually might be happening (and I doubt it, but if there is a small chance, we should look into it) I am going to add timestamping to the bug-tracker so we (you, me, and anyone else on the development team) can cross-reference such events. It sounds an awful lot like #1 is the case, and if we could peek at the PhotoLog table and discover that, yes indeedy, Jay managed to push two photos in a row from 4 votes to 5 votes, thereby knocking them into smoothed territory. Out of curiosity, why did you choose a threshhold of 5? The conditions a photo must satisfy to make it onto the front page are that it must have 3 votes, and be marked as allow_votes='Y' (so that people cannot gerrymander a bunch of votes and then freeze the rating -- you freeze the rating, you get knocked off the FP rotation, simple as that!). So if the algorithm worked for N=3 votes that would be a logical threshhold. Also, if the algorithm itself is prone to sabotage, I can suggest a number of other, somewhat more sophisticated pattern recognition algorithms that will reliably cluster around a 'true' global perturbation minimum without aberrant results such as Jay witnessed. Just a thought, and not necessarily a very useful one.
|
|
|
|
|
biff
Jul 20, 2003, 7:53 PM
Post #21 of 23
(3661 views)
Shortcut
Registered: Nov 5, 2001
Posts: 851
|
In reply to: Out of curiosity, why did you choose a threshhold of 5? The conditions a photo must satisfy to make it onto the front page are that it must have 3 votes, and be marked as allow_votes='Y' (so that people cannot gerrymander a bunch of votes and then freeze the rating -- you freeze the rating, you get knocked off the FP rotation, simple as that!). So if the algorithm worked for N=3 votes that would be a logical threshhold. Also, if the algorithm itself is prone to sabotage, I can suggest a number of other, somewhat more sophisticated pattern recognition algorithms that will reliably cluster around a 'true' global perturbation minimum without aberrant results such as Jay witnessed. Just a thought, and not necessarily a very useful one. I chose 4or5 becuse the algorithm is based on statistics .. and statistics for small numbers are useless. is good enough for the front page, because when something hits the front page it typically gets atleast 2 more votes, which will then be enough to get some statistics from. The algorithm isn't prone to sabotage, but if somone wanted to maximize the deflating or inflating effect of their vote, knoledge of the algorithm would make it a little easier. I believe some people have already figured out how to vote to have the biggest effect, but since I implemented the algorithm, most voters have become good citezens and voted on the photo, rather than on the user who submitted it .. In recent months I have not noticed very many sabotage votes in the system.
|
|
|
|
|
jt512
Jul 21, 2003, 5:19 PM
Post #22 of 23
(3661 views)
Shortcut
Registered: Apr 12, 2001
Posts: 21904
|
In reply to: ...since I implemented the algorithm, most voters have become good citezens and voted on the photo, rather than on the user who submitted it .. In recent months I have not noticed very many sabotage votes in the system. I think your imagination is getting the better of you. I doubt that many users were aware that any kind of algorithm had been implemented or have changed their voting behavior. -Jay
|
|
|
|
|
rrrADAM
Jul 21, 2003, 6:43 PM
Post #23 of 23
(3661 views)
Shortcut
Registered: Dec 19, 1999
Posts: 17553
|
It should be 5 anyway... But that is just MHO.
|
|
|
|
|
|