Scores load much lower in updated client.

Case number:699969-992542
Opened by:smilingone
Opened on:Tuesday, May 1, 2012 - 05:45
Last modified:Wednesday, July 25, 2012 - 07:32

A saved solution of 9944 loads as 9919 in the updated client, but, loads fine as 9944 in a NON updated client. What's up with that? I'm glad I only updated 1 client thus far. I run them all in their own folder, so, my others aren't affected.

(Tue, 05/01/2012 - 05:45  |  21 comments)

Joined: 05/09/2008
Groups: None

so it loads as 9919 in the 'devprev' client, but 9944 in the 'main' client?
Does your solution contain an pi-helix?

Joined: 09/19/2011
Groups: Beta Folders

I just know what I like the look of and which recipes appear to be productive. The science is a complete unknown to me, including most of the related terms.

Odds are, there isn't since it is a relatively low score for 550 in comparison to others. I've heard these pi-helixes create a higher score. Do you have a link showing what one looks like?

Joined: 05/09/2008
Groups: None

Here's a link to a pi-helix:

The issue you bring up "I just know...which recipes appear to be productive" is what this update is trying to fix.

There was a bug in Rosetta that scores these pi helices way too much, and this was discovered by Foldit player recipes that highlighted this error.

We fixed the bug, and this update now corrects the error.

Right now it's only in "devprev", but soon (tomorrow I believe) it will be in "main" and everyone will have the exact same score function and everyone's solution will load the same way.

This may cause a few recipes to be less productive, however!

Joined: 09/19/2011
Groups: Beta Folders

Why on earth would you guys provide a fix that doesn't roll out for all at once? Talk about unfair. I'm sure you have your reasons, but, it doesn't make sense to me. None of the clients should work without this fix if that is the case. JMO

Btw, thanks for the link. ;)

Joined: 06/17/2010

This update is NOT optional.
It will go on main in 24hrs.
Then all scores from NOT updated client will NOT count!

Joined: 09/19/2011
Groups: Beta Folders

I hope that is the way it works... that's the only fair thing to do. I would like to know the exact time this occurs so that I can stop all recipes and update immediately.

brow42's picture
User offline. Last seen 19 weeks 9 hours ago. Offline
Joined: 09/19/2011
Groups: None

They're just making sure it doesn't crash, or that all helices score -1000 now, or something like that. If they did, you can be sure phi16 would give them an earful! The original plan was to shut down foldit for a day and make all puzzles end on the same day! I don't think either the players or the devs liked that much, so that's why we have the score glitch today.

Joined: 05/09/2008
Groups: None

Exactly... we appreciate everyone using "devprev" for today to make sure everything doesn't descend into chaos when we update this to "main" as well ;-)

If you can't stand having lower scores than other players today, feel free to switch from "devprev" to "main" but by tonight everyone will be on the same scoring system.

This is why we waited to do this until 548 closed yesterday, so that there would be no puzzles closing in the next 2 days.

Joined: 09/19/2011
Groups: Beta Folders

I know you are doing your best, but, I'm also assuming you want to know how this new rollout is working.

Just picked up a share with an unupdated client and didn't even have to evo it. It was created on an updated client and loaded much higher on the NON updated client. lol Will the scores on the current scoreboard also be wiped once all scoring is the same? If not, there is a huge scoring mess and none of the scores from the currently running puzzles will have any real value.

When the reverse happened and I ended up with a lower score, the scoreboard still showed the higher score.

Joined: 09/19/2011
Groups: Beta Folders

I haven't a clue what a pi-helix is.

Joined: 08/30/2011

i'm not completely sure what it is either but as I understand it, it is like a helix but with a bigger diameter. And this kind of helix was scoring to good and people knew that and used it to make good scores and produced solutions which are high scoring but useless.

Joined: 09/19/2011
Groups: Beta Folders

Just an fyi, my solution doesn't even have a helix of any sort in it.

The entire game should have been shut down and everyone updated at the same time.

Joined: 05/09/2008
Groups: None

To do that we would have had to either:

1) Have all the puzzles close at the same time (which nobody wants)
2) Not post any new puzzles until the last one closed (so for 1 week you'd have very few puzzles, and for a few days there would only be one puzzle up)

Again, everyone will be in the same boat tonight.

Alternatively we could make all the currently open puzzles worth 0 points.

I will not post the first CASP 10 puzzle until the new client gets posted to "main" so that everybody will have the exact same score function for the duration of that (and future) puzzles.

We do apologize for this, but we really had to post this fix (especially before CASP started), and we tried to warn everyone about it:

Thanks for your patience... I only wish we had dealt with this way back in the day when the pi-helix loophole was first uncovered by you guys.

Joined: 09/19/2011
Groups: Beta Folders

It isn't only affecting pi helix scoring. The puzzle I first noticed the drop in score in doesn't have any helixes.

Joined: 05/09/2008
Groups: None

The only change is for "n to n+5" hydrogen bonds (as infjamc mentions below) so it will also alter the score of ANY structure that forms a backbone hydrogen bond with a residue 5 spots before (or after) it, even if it's not in a helix (such as a turn in a loop or something).

I know this is frustrating, I know it's a pain, but the Baker Lab has thoroughly benchmarked this score change (which is why it took this long to roll out to Foldit) and the new scoring is the correct scoring and how it should have behaved all along.

You all know that the Rosetta Energy Score function is far from perfect, and there are many people in the Baker Lab working on improving it, but this single change is at least going in the right direction and fixing one element that was wrong.

Joined: 09/19/2011
Groups: Beta Folders

What time will the final changes be made? I'd like to have my recipes stopped.

And, thanks for all you do. I'm sure most, if not everyone appreciates all you do. ;)

Joined: 05/09/2008
Groups: None

jflat said the final update should be posted around 4pm PDT (23:00 GMT)

I will let him know that he should hop into chat to warn everyone first!

Thanks again for being so understanding... hopefully we can quickly pull the band-aid off and not drag this out any longer!

infjamc's picture
User offline. Last seen 3 hours 32 min ago. Offline
Joined: 02/20/2009
Groups: Contenders

Here you go:

To make a long story short, it's a helix whose hydrogen bond pattern is "n to n+5" instead of "n to n+4" (as in the standard alpha helix). Such a structure would score unreasonably high previously because the hydrogen bonds in a pi helix would be mis-categorized under the Rosetta scoring function.

mimi's picture
User offline. Last seen 1 year 32 weeks ago. Offline
Joined: 11/17/2008
Groups: Contenders

I've just had the same thing loading an evo for 550 on the updated DevPrev client.
Evo score is 10437 but loads as 10424 while the requirement for evolving stays at 10439

Structure does not have any helices at all

spmm's picture
User offline. Last seen 1 hour 15 min ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

perhaps we should just let the update happen and then discuss remaining score issues. There is presumably a necessary hiatus between devprev and main updates.
Evo scores may be affected or it may be a separate issue. Mike has posted on evo score problems here;

Joined: 09/19/2011
Groups: Beta Folders
Status: Open » Closed



Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, RosettaCommons