Scoring anomaly Hydrogen Bond Network 2

Case number:954892-2000667
Topic:Developer Preview
Opened by:Bletchley Park
Status:Open
Type:Bug
Opened on:Friday, May 1, 2015 - 12:34
Last modified:Thursday, May 28, 2015 - 00:02

I ran a script (indicated in the solution), it produced a higher score, but none of the 'restore score'options, nor the history showed the higher score I saw recorded on the server.

Uploaded solution as BP scoring anomaly.

(Fri, 05/01/2015 - 12:34  |  32 comments)


Joined: 05/19/2009
Groups: Contenders

Update: I have had the situation again and this time I saw the score on the screen, as it correctly recorded on the server (130101). However, it was registered as 'credit best'. A lower score was and again I could not reload the *verfied* highests score by restoreign credit best, absolute best etc.

Joined: 05/19/2009
Groups: Contenders

typos...
1. score was 13101.
2. "However, it was NOT registered as çredit best' ...".

Joined: 05/19/2009
Groups: Contenders

situation:
- restore abs best,
- 2834.5 bonus, score = 13035.001
- wiggle
- 2745.6 bonus, score = 13079.858
- restore abs best
- score returns to the best bonus, not to the highest score.

Both ctrl-B and I have the same score of 13035, whereas 13101 is recorded (correctly, I observed it earlier as it was reached) on the server.

Madde's picture
User offline. Last seen 5 weeks 5 days ago. Offline
Joined: 05/29/2008
Groups: Void Crushers

Same here. Server says my highscore is 16061, my autosaved very and credit bests say the same but they load as 15981.
Autosaved credit best shared with Scientists.

Joined: 06/06/2013
Groups: Gargleblasters

I have had this problem with other puzzles (e.g. 1080) even before this. I killed a sript mid-way, and assumed that was the problem

Joined: 06/06/2013
Groups: Gargleblasters

I find the scoring on this hard to understand. My best score had only one or no H bond networks. I just mutated using two scripts. As soon as I started to try to get h bonds, my overall score has gone down
I am the first to admit I am not good at mutation. However, I don't think this puzzle is ready for main if the goal is to get H-bonds and the scores of an average person like me seem to penalize them. I think a lot of players will be discouraged.
I do really like the thunderbolts so you can see. But I have no clue why my total score is so bad as my h bonds increase

Joined: 07/21/2013
Groups: Beta Folders

I put in a bug report about a very large score necessary for evolving. I shared with the Scientists the three shares that had this characteristic. Jflat6 came up with the idea that these shares had originally been created with an older version of Foldit and this was the problem. Could you share with Scientists the puzzle shares that have this problem and perhaps JFlat6 could look at them too?

Joined: 05/19/2009
Groups: Contenders

Hi brgreening, I posted examples to scientists for all the issues I reported as I posted the feedbacks.

Madde's picture
User offline. Last seen 5 weeks 5 days ago. Offline
Joined: 05/29/2008
Groups: Void Crushers

Still happening in p1087.
Credited score on the server: 8878 - Both Autosaved Very Best and Credit Best say Score 8878 but load as 8805. Shared my Autosaved Very Best with scientists.
This time no script was involved.
Build-ID: 20150512-0e5b7a4ade-win_x86-devprev

jflat06's picture
User offline. Last seen 1 day 22 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

That build-id appears to be from the old devprev. That version will still have the issue. Can you give me the full hash of the binary? I can blacklist that client from reporting scores to the server, to prevent any abuse. I will also need the hashes of the linux/mac versions, if anyone has those.

Madde's picture
User offline. Last seen 5 weeks 5 days ago. Offline
Joined: 05/29/2008
Groups: Void Crushers

5e4e06d735fc4da0396237b852663b10

That's in the version-binary.txt file. Is that what you were asking for?

Edit:
I just updated Foldit on my other computer and it turned out that 20150512-0e5b7a4ade-win_x86-devprev is still the newest devprev on the server. (13:20 UTC)

Joined: 09/24/2012
Groups: Go Science

for a mac:
version binary: 58bc21651a4686c3f84c4a2db7c33e88

This is the main that I downloaded on 2 days ago, when starting 1087

Joined: 09/24/2012
Groups: Go Science

for a pc:
5e4e06d735fc4da0396237b852663b10

Joined: 09/24/2012
Groups: Go Science

buildid: 20150512-0e5b7a4ade-win_x86

This is on MAIN !

Joined: 07/21/2013
Groups: Beta Folders

Bruno, could you mention the name of the file and where it can be downloaded from that produces hashes on pc's?

thx in advance.

Joined: 05/19/2009
Groups: Contenders

To what client version should we move to keep getting our scores registered ?
We have a puzzle ending tonight.

Indeed that 20150512 is the latest version I downlaoded today.

Joined: 09/24/2012
Groups: Go Science

Now in main client, we have the same problem with 1087.

For example, all restores give 10200
During a recipe, we see the score going up to 10300, and it's credited 10300 on the server.
Undo shows 10300.
Clicking on undo, or recent best or credit best or absolute best loads with 10200.
get score within recipe gives 10200.
Best quick saves and autosaves give 10200.

BUT when sharing: needs 10302 to evolve. => quite impossible to evolve high scores.

I'm not the only one observing this in my team.

All clients are main and latest update.

Angus's picture
User offline. Last seen 5 days 4 hours ago. Offline
Joined: 06/04/2008
Groups: Beta Folders

I agree, the issue is still there on 1087. I shared to scientists one that was also exactly 100 points too high on the scoreboard. 14119 loads as 14019 .

spmm's picture
User offline. Last seen 27 weeks 3 days ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

Yes me three on 1087- restore credit best is not restoring my high score on the server. No cut point scripts used. 4 days 15 hours to run. Mac 10xx latest deprive.

Joined: 09/24/2012
Groups: Go Science

Tip to have a chance to keep the credit highest result:

after stopping a recipe
-recent best gives the highest result.
-credited best and absolute best give a lower result ! AND we definitively loose the higher result

If starting with recent best, credited best and absolute best have a chance to be kept and solving the problem above

Joined: 03/05/2015
Groups: Gargleblasters
Type: DevPrev Issue » Question

I'm also experiencing scoring anomalies (lost most of a 4500 pt. bonus on a solution over 15,400 and have not been able to recover it to anything more than 12,200 with a declining bonus every time I try to work it, despite recreating the sequence AA for AA and shaking the daylights out of it). Quickly reading through this thread I will try closing foldit and downloading the latest build and using recent best after the recipe if I experience the problem again.

I agree this one is not ready for prime time and as a relatively new player it is very disheartening to see points evaporate for what seems no good reason.

Joined: 03/05/2015
Groups: Gargleblasters

I'm also experiencing scoring anomalies (lost most of a 4500 pt. bonus on a solution over 15,400 and have not been able to recover it to anything more than 12,200 with a declining bonus every time I try to work it, despite recreating the sequence AA for AA and shaking the daylights out of it). Quickly reading through this thread I will try closing foldit and downloading the latest build and using recent best after the recipe if I experience the problem again.

I agree this one is not ready for prime time and as a relatively new player it is very disheartening to see points evaporate for what seems no good reason.

Joined: 09/21/2011
Groups: Void Crushers

Latest updates dont help....

Joined: 04/19/2011

I have the same problem as everybody here and after going through all comments, still no solution. Bruno's suggestion using the recent best does not work for me. Recent best doesn't load the recent best, just some random low score. Interestingly the autosaved credited- or very best displays the score right, but when I am loading any of that, always loosing a couple 100 points. Very hard to improve the score since every time I stop, have to start again from at least a 100 points lower score...
Evolving a high scoring solution even harder, if not impossible as Bruno mentioned it earlier

Seems to me there is some bad scoring algorithm associated with this puzzle. (I'm using the latest client).
Is it possible to have a fix during this run (2 days left) or the puzzle will be re-posted after correction?

Joined: 04/24/2014
Groups: None

Regarding 1087, a repost is coming: http://fold.it/portal/node/2000741

jflat06's picture
User offline. Last seen 1 day 22 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

If anyone can find a way to reliably reproduce this, that would be very helpful.

I am able to see the shared solutions that load in lower than their score indicates. What I need is to be able to see the state of the program/scoring/hydrogen bonds when it records the high score. This should show why the scoring anomaly is taking place, and I should be able to fix it. I'm not sure this is possible, since this seems to be intermittently reproducible.

Madde's picture
User offline. Last seen 5 weeks 5 days ago. Offline
Joined: 05/29/2008
Groups: Void Crushers

I just shared a solution (1087-1s2) with scientists on p1087 with instructions how to reproduce this.

jflat06's picture
User offline. Last seen 1 day 22 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

Thanks... having trouble with this currently, because apparently this script freezes up the Foldit GUI on Linux. Not sure why that's happening.

Joined: 05/19/2009
Groups: Contenders
Type: Question » Bug

Would a breakpoint on score changes that trigger a server update help ?

jflat06's picture
User offline. Last seen 1 day 22 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

Unfortunately, probably not. You would have to have some way to detect when the problem is happening, and break only on that. You would never find it just breaking every time the score changes, since that happens many times per second, and this bug is highly sporadic. That's part of why we need a reproducible case so badly.

Detecting when the problem is happening is tricky, since that requires knowledge of why the problem is happening to begin with. All we know is that certain solutions are scored higher initially compared to when they are re-loaded in, and that isn't something that can be tested every time the score changes.

jflat06's picture
User offline. Last seen 1 day 22 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

If anyone has an example of this happening in another puzzle (I recall some players mentioning this), please share that solution with scientists and mention the puzzle/solution here!

Joined: 05/19/2009
Groups: Contenders

I sent you a PM with some suggestions that may help yesterday.

Sitemap

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