Movie already see
Moderators: Stardust@home Team, DustMods
Movie already see
Maybe someone have already write about this thing, but I would ask you why sometimes I see the same movie (no a test movie, but a real movie).
This fact cause a few time lag in the search.
This fact cause a few time lag in the search.
Re: Movie already see
Interesting question, unofficial (but I hope helpful) answer:goldrake wrote:Maybe someone have already write about this thing, but I would ask you why sometimes I see the same movie (no a test movie, but a real movie).
This fact cause a few time lag in the search.
The analysis is akin to the "bar bet" relating to "how many people present have the same birthday (month and day)". The number present is (to many people) surprisingly low (in the 20's IIRC) to reach "break even" odds.
So I plugged some assumptions into a spreadsheet and let the computer compute the odds that one specific person would see the same movie more than once. Then expanded to compute the odds that at least one reviewer of the 10,000+ would see the same move more than once.
Assume 1,000,000 movies in pool.
Assume 10,000 reviewers, each reviewing at least 500 movies.
The probability that a specific reviewer would have a duplicate during the first 500 movies is small, about 0.1 percent (P=0.001)
The probability that at least one reviewer (of the 10,000) would view a movie twice is very high, about 99.5%!
Overall the lost productivity of re-reviewing a movie is a very small percent.
On the other side of the equation, the cost to check if you have previously reviewed the movie that StarDust has randomly selected is nontrivial, and gets increasingly expensive as the number of movies that you have reviewed increases.
So, their "imperfect scheme" appears to me to be the "best scheme."
The surprizing thing to me is that you recall having seen that specific movie before!
Re: Movie already see
Same thing happened to me with movie 5509613V1goldrake wrote:Maybe someone have already write about this thing, but I would ask you why sometimes I see the same movie (no a test movie, but a real movie).
This fact cause a few time lag in the search.
My total movies viewed is 3700
Clicked on it both times and it created duplicate entry in My Events
Looks like a software bug (It ought to check to see if the movie is already in the list before inserting)
I am thinking of giving up for a week or two until they have scanned more tiles
Re: Movie already see
StanPope wrote:....
So, their "imperfect scheme" appears to me to be the "best scheme."
The surprizing thing to me is that you recall having seen that specific movie before!
Thank you for information.
I remember the specific movie because I've seen this movie only 5 movies earlier and the movie was enough characteristic.
My total movies viewed are 7759.
I've meet this problem at least 3 time or more.
Your theory is correct for me, but there is something wrong.
Re: Movie already see
This is true if it stores a list of previously viewed movies for each volunteer - but there are other possible strategies eg.StanPope wrote: ........
On the other side of the equation, the cost to check if you have previously reviewed the movie that StarDust has randomly selected is nontrivial, and gets increasingly expensive as the number of movies that you have reviewed increases.
So, their "imperfect scheme" appears to me to be the "best scheme."
The surprizing thing to me is that you recall having seen that specific movie before!
Check the number of viewed movies value for the volunteer and if its more than say 2000 then choose a movie that has been scanned more recently. This would give the heavy workers a higher % of fresh movies
Re: Movie already see
I don't think we want to go there ... that compounds the probability of garnering fresh movies above the natural advantage from looking at more movies... to the detriment of the more casual participants.merelake wrote:other possible strategies eg.
Check the number of movies value for the volunteer and if its more than say 2000 then choose a movie that has been scanned more recently. This would give the heavy workers a higher % of fresh movies
But you are right that there may be viable alternative strategies that would not bring the server to its knees. I just don't know what they would be.
In the computation, I assumed a value for the number of movies in the pool to be 1,000,000. I think that is very high. With fewer movies, the probability of repeats increases, of course.
I think that the number of volunteers is currently much greater than the job requires ... 100+ reviews of a movie seems dreadful overkill! I just hope that enough of them stay around for when the workload gets up to volume! And, while overreviewing, their skills continue to improve.
Re: Movie already see
Yes I agree it is certainly overkill.StanPope wrote: I think that the number of volunteers is currently much greater than the job requires ... 100+ reviews of a movie seems dreadful overkill! I just hope that enough of them stay around for when the workload gets up to volume! And, while overreviewing, their skills continue to improve.
Also there is a bias caused by the fact that the first few tiles have spent more time on the server so they will end up with more viewings.
Of course it will average out if the project goes on for an infinite time..............................
Another consequence of the random method is that a few movies will get zero viewings and a few will get 1000 viewings.
I would hope that they would store the number of viewings for each movie and retire them when they get past a certain point.
Today I've viewed two identical real movies with a distance from the first to the second of only 10 movies
My surprise is to see that the ID number was different: 1982792V1 and 5527523V1
Take a look:
http://stardustathome.ssl.berkeley.edu/ ... =1982792V1
but if I recall the second ID I don't see the same movie.
http://stardustathome.ssl.berkeley.edu/ ... =5527523V1
Nevertheless if you go at
http://stardustathome.ssl.berkeley.edu/ ... rds=?-1,-1
the ID is 5527523V1 and is the same movie of ID 1982792V1 .
My surprise is to see that the ID number was different: 1982792V1 and 5527523V1
Take a look:
http://stardustathome.ssl.berkeley.edu/ ... =1982792V1
but if I recall the second ID I don't see the same movie.
http://stardustathome.ssl.berkeley.edu/ ... =5527523V1
Nevertheless if you go at
http://stardustathome.ssl.berkeley.edu/ ... rds=?-1,-1
the ID is 5527523V1 and is the same movie of ID 1982792V1 .
-
- Posts: 47
- Joined: Wed Aug 09, 2006 11:38 pm
It's sort of two movies and sort of three movies. Your first two links are simple links to two distinct real movies. But that third link is a live link (which should not normally be used) to a calibration movie. Calibration movies are presented with the ID of a real movie, although that ID is quite unrelated to the movie shown. It happens that this calibration movie is actually the real movie with the other ID.goldrake wrote:Today I've viewed two identical real movies with a distance from the first to the second of only 10 movies
My surprise is to see that the ID number was different: 1982792V1 and 5527523V1
Take a look:
http://stardustathome.ssl.berkeley.edu/ ... =1982792V1
but if I recall the second ID I don't see the same movie.
http://stardustathome.ssl.berkeley.edu/ ... =5527523V1
Nevertheless if you go at
http://stardustathome.ssl.berkeley.edu/ ... rds=?-1,-1
the ID is 5527523V1 and is the same movie of ID 1982792V1 .
Tom Gutman
Ok, thank you, but the movie http://stardustathome.ssl.berkeley.edu/ ... =1982792V1 is not a calibration movie.Tom_Gutman wrote:It's sort of two movies and sort of three movies. Your first two links are simple links to two distinct real movies. But that third link is a live link (which should not normally be used) to a calibration movie. Calibration movies are presented with the ID of a real movie, although that ID is quite unrelated to the movie shown. It happens that this calibration movie is actually the real movie with the other ID.goldrake wrote:Today I've viewed two identical real movies with a distance from the first to the second of only 10 movies
My surprise is to see that the ID number was different: 1982792V1 and 5527523V1
Take a look:
http://stardustathome.ssl.berkeley.edu/ ... =1982792V1
but if I recall the second ID I don't see the same movie.
http://stardustathome.ssl.berkeley.edu/ ... =5527523V1
Nevertheless if you go at
http://stardustathome.ssl.berkeley.edu/ ... rds=?-1,-1
the ID is 5527523V1 and is the same movie of ID 1982792V1 .
Why I've seen two times during my normal searching session?
-
- DustMod
- Posts: 694
- Joined: Wed May 17, 2006 8:12 pm
- Location: Horsetown, USA
Thank you.DustBuster wrote:The movies are in a pool that are served up randomly. There is a chance you will see the same movie again (I have seem many movies more than once).goldrake wrote:Why I've seen two times during my normal searching session?
Does that help?
I know that some other users have noted this problem, but I think that the probability is very low, so I'm afraid that the method for the random distribution of the movies is not so good.
However this little problem don't have compromise my results for now, or at least I hope.
At random means at random. Even when the pool is large, an unlikely event (twice the same result in a row) may occur.
When the number of events is high (as it is here with many users each drawing multiple events) the appearance of the unlikely event even becomes likely.
Only for the (un)fortunate user who encounters the unlikely event it appears as strange.
Makes sense..??
When the number of events is high (as it is here with many users each drawing multiple events) the appearance of the unlikely event even becomes likely.
Only for the (un)fortunate user who encounters the unlikely event it appears as strange.
Makes sense..??
Just dusting...
-
- Posts: 47
- Joined: Wed Aug 09, 2006 11:38 pm
Nothing shown by the event viewer is a calibration movie. That form of URL can only bring up real movies. Calibration movies show the same ID as real movies, even though that ID is not actually related to the calibration movie. Using the event viewer on the ID of a calibration movie brings up the real movie with that ID. Only using the VM URL will bring up the calibration movie.goldrake wrote:Ok, thank you, but the movie http://stardustathome.ssl.berkeley.edu/ ... =1982792V1 is not a calibration movie.
Why I've seen two times during my normal searching session?
You have three URLs. The two event viewer URLs bring up two distinct real movies, each with its ID. The VM URL brings up a calibration movie. This calibration movie has the ID of one of those two real movies, and the content of the other one.
Tom Gutman