Incorrect PSI generation

Encounter an error, or something which isn't working correctly? Please, let us know
djross
Posts: 1212
5318 Ratings
Your TCI: na
Joined: Sun Apr 16, 2006 12:56 am

Incorrect PSI generation

Post by djross »

I thought I should open a separate bug report for this, which I reported in the thread announcing the new percentile algorithm.

djross wrote:
AFlickering wrote:
djross wrote:PSIs are definitely being incorrectly calculated at the moment, because there is a mix of old TCIs (with lower numbers) and new TCIs (which are all higher numbers). So, when I look at the PSI for Tarkovsky's Mirror, for example, all ten numbers used to generate the PSI are from those old TCIs. It is only further down the list, after these ten, that my actually closest users appear. Clicking the "see why" button that shows how the score was calculated confirms that none of my actually closest users factored into the PSI.


i've not had this issue at any point for the record, all my TCIs converted when i made the switch.


It is no longer true for the example I gave, and I have not found another example, so perhaps it is no longer an issue.


This is again true, including for the example I gave, Tarkovsky's Mirror, and for many other movies. Using that example, the PSI generated on the film page is calculated on the basis of users with TCIs ranging from 1.2169 to 1.7635 (in other words, TCIs that are being calculated according to the "tier system" rather than the "percentile system"). These users include some of my closest users, but not all. For example, it does not include user lisa-, because for some reason this user has a TCI calculated by the new system, which is 15.2325, and so far below the ten used to calculate the prediction for this film. In reality, lisa- is one of the users with the closest TCI to me, and should be included in the calculation of the PSI for this film.

Exactly the same thing is true of Taxi Driver, for example, this time excluding both lisa- and dardan, another user who is one of my closest in terms of TCI and should be included in the calculation for that film.

An odd thing about it is that even a user whom I know has switched to the percentile system, such as jgreenwood, is having her TCI calculated in the old way. But then, if I go to her user page, the TCI will be recalculated correctly, with the result that she too will be excluded from PSI calculations from all those films because she no longer has the low number associated with the old system.

In short, predictions are not currently being generated on the basis of the closest users.

mpowell
Posts: 3851
1201 Ratings
Your TCI: na
Joined: Fri Sep 09, 2005 10:22 am

Re: Incorrect PSI generation

Post by mpowell »

Alright, thanks for reporting this. We'll re-open the bug that we closed, and take another look. We'll get this one squashed soon!

mpowell
Posts: 3851
1201 Ratings
Your TCI: na
Joined: Fri Sep 09, 2005 10:22 am

Re: Incorrect PSI generation

Post by mpowell »

Alright guys, I think NOW we've fixed this problem, of "tier-based" TCIs entering the system even though you've switched to percentiles. It was a tough one to track down, but we've been monitoring now for a couple days, and think it looks better.

If it happens again, please let us know. But before you do, please send us a bottle of whiskey, because otherwise we might put our heads through a wall.

livelove
Posts: 759
67 Ratings
Your TCI: na
Joined: Thu Sep 17, 2015 9:36 pm

Re: Incorrect PSI generation

Post by livelove »

mpowell wrote:[…] problem, of "tier-based" TCIs entering the system even though you've switched to percentiles. It was a tough one to track down […] we've been monitoring now for a couple days […] If it happens again […] we might put our heads through a wall.

:| Forgive me for saying so, but it appears as though you were paying a high price for maintaining 2 systems running in parallel under the hood, imaginably binding resources then unavailable elsewhere.

Post Reply