Page 5 of 9

Posted: Mon Aug 07, 2006 9:09 am
by DustBuster
pointyhat wrote: Surely movies should not reviewed more than once by the same person?
I don't know if that happens or not, but it would make sense to calibrate sensitivity partially by checking if someone clicked 'track' one time and then 'no track' on review of the same movie. But that doesn't help explain why it would show up more than once in myevents.

Posted: Mon Aug 07, 2006 9:46 am
by pointyhat
Maybe it`s a bug. And what should be the problem? "Nice to see you again, little grain
Well I guess the problem could be that by virtue of me having viewed enough movies to get the same movie randomly selected twice my opinion on it is getting double weighted compared to those who have viewed it only once, (if indeed it is being randomly selected rather than a specific mechanism to help calibrate sensitivity). Now I guess one could argue that by virtue of viewing lots of movies participants get better at reading potential signal from noise (and indeed there are some movies I marked in the first couple of hundred I viewed as containing a track that I would now mark as no track), however until enough time has passed on the project for the Stardust team to be able to indicate how participants are doing via the my events list (which very understandably is going to take a while), I have no reason to believe that I am in fact getting better at identifying movies that the Stardust team would be interested in since I can spot a calibration movie at one hundred yards.


Gavin

Posted: Mon Aug 07, 2006 10:46 am
by stardust1
pointyhat wrote:
Maybe it`s a bug. And what should be the problem? "Nice to see you again, little grain
Well I guess the problem could be that by virtue of me having viewed enough movies to get the same movie randomly selected twice my opinion on it is getting double weighted compared to those who have viewed it only once
d´accord. If I´ll get one more candidate a second time, I will verify if my second click will influence its "number of agreements"
I have no reason to believe that I am in fact getting better at identifying movies that the Stardust team would be interested in since I can spot a calibration movie at one hundred yards.
[/quote]

I have reason to believe this. My last "incorrectly" was about 2000 movies ago and was one of those buggy ones (flagged as incorrect with a track of 30 microns in it - there are about 12 of them, I got them all, some of them twice, dropping down my sensitivity to a - therefore statistically fixed - value of 98-99%).
That´s boring by and by (not only to me) and a waste of time. You will get to the same point soon. So in my opinion training on real movies would improove sensitivity.

regards

Peter (stardust1)

Posted: Mon Aug 07, 2006 10:57 am
by Orion_0169
Stardust,

I'm in the same boat. Callibration counts are so high that the percentages can barely budge. Also same with errors, I've only had two or three in the last 1000+ callibration movies and those are all the buggy ones. Mind you, it APPEARS that some of those buggy ones might not be around anymore. Or perhaps the frequency of showing them has dropped. I haven't seen as many of those ones as I used to. Not even close. Particularly my nemesis 11796. 4x. Argh!

Posted: Mon Aug 07, 2006 11:03 am
by pointyhat
If I´ll get one more candidate a second time, I will verify if my second click will influence its "number of agreements
I believe it does, movie in question had 10 confirmations before I reconfirmed it, then it had 11.

All the best
Gavin

Posted: Mon Aug 07, 2006 11:07 am
by stardust1
pointyhat wrote:

Code: Select all

If I´ll get one more candidate a second time, I will verify if my second click will influence its "number of agreements[quote]

I believe it does, movie in question had 10 confirmations before I reconfirmed it, then it had 11.

All the best
Gavin[/quote]
So it is a bug!

Peter (stardust1)

Posted: Mon Aug 07, 2006 11:09 am
by Orion_0169
Talked about that elsewhere a bunch. Having you checked against yourself doesn't make it a bug. It pads the stats though.

Posted: Mon Aug 07, 2006 11:25 am
by css94381672
Hi everybody !

Isn´t it a little silly, that I am able with 100% fruitfulness make a prediction whether a coming movie will be testing or real?

e.g.
http://stardustathome.ssl.berkeley.edu/ ... pe=testing[/b]&code=&coords=?-1,-1

(P.S. from Czech Republic, no english speaking so be benevolent to my english :oops:

Posted: Mon Aug 07, 2006 11:36 am
by Orion_0169
css Just don't let it become a habit. First off there are callibration movies with VERY small tracks that are easy to miss. Also some of the features that allow you to quickly identify callibration movies sometimes show up in regular movies.

Posted: Mon Aug 07, 2006 11:43 am
by stardust1
Orion_0169 wrote:Stardust,

I'm in the same boat. Callibration counts are so high that the percentages can barely budge. Also same with errors, I've only had two or three in the last 1000+ callibration movies and those are all the buggy ones. Mind you, it APPEARS that some of those buggy ones might not be around anymore. Or perhaps the frequency of showing them has dropped. I haven't seen as many of those ones as I used to. Not even close. Particularly my nemesis 11796. 4x. Argh!
I think some of those buggy ones had been eleminated yesterday (same time when the movie-IDs shown on the page changed).
And: it seems to me as if the frequency off all calibration movies has dropped. Maybe only to me and you and other highscorers, which would be pretty fair to all those other enthusiasts on board. If so: I´m going to enjoy viewing more real ones.

11796 was new to me or I didn´t care

[edit]: 11796 ist still "on the wild" and names now "Movie id: 7194462V1" (if not randomly generated) [/edit]
[edit2] 9902: my second found wrongly flagged / missleading "no-track" calibration movie (containing very small track or whatever) [edit2]

kind regards

Peter (stardust1)

Posted: Mon Aug 07, 2006 12:25 pm
by Usher73
Another issue #3, two movies in one, ID 2290703V1

Posted: Mon Aug 07, 2006 1:27 pm
by ToSeek
css94381672 wrote:Hi everybody !

Isn´t it a little silly, that I am able with 100% fruitfulness make a prediction whether a coming movie will be testing or real?

e.g.
http://stardustathome.ssl.berkeley.edu/ ... pe=testing[/b]&code=&coords=?-1,-1

(P.S. from Czech Republic, no english speaking so be benevolent to my english :oops:
From what I can tell, it's not always right, however.

Posted: Mon Aug 07, 2006 5:52 pm
by gentboy
I don't know if anybody else met this problem, but I didn't find it in FAQ:
After I focus on some other window except the VM and back to the VM, the VM seemed no response to my zoom action.

Posted: Tue Aug 08, 2006 3:22 am
by Galathea
I have a strange problem: every time I click on a calibration movie with a track, the VM considers that I clicked on that one and automatically the next one as well, that I can just glimpse before it disappears. I have tried to click very gently but it does not change a thing: it just happened three times in a row, right now.

I am not too interested in my ranking, but it adds a lot of noise to my list of events, as I seem to find a lot of dust where there is obviously none...

Posted: Tue Aug 08, 2006 4:36 am
by css94381672
Orion_0169 wrote:css Just don't let it become a habit. First off there are callibration movies with VERY small tracks that are easy to miss. Also some of the features that allow you to quickly identify callibration movies sometimes show up in regular movies.
I know, but when it is predefined in url, some script sould do it for me...
ToSeek wrote:From what I can tell, it's not always right, however.
yes? ok, I tried it only few times
(but still shinking scientists should reslve it)


(P.S. let's our forum colorful :wink: )