Evolver scoring

Case number:671071-812414
Topic:Game: Other
Opened by:DisposableHeart
Opened on:Thursday, January 8, 2009 - 15:19
Last modified:Tuesday, June 14, 2011 - 22:14

The current evolver scoring does not support the idea of finding the best overall solution.

While group solutions are better -in general- than solo solutions they tend to evolve the first submitted solo solution that might not be the optimal one. Unless another solo is taking over, not even just getting close to the current best group solve, however it would yield much better result in group, buried under the rush for a better evolver ranking.

Please consider.

(Thu, 01/08/2009 - 15:19  |  4 comments)

jas0501's picture
User offline. Last seen 1 year 47 weeks ago. Offline
Joined: 12/08/2008
Groups: None

Do you have an alternate approach?

Entering Blue sky mode...

One thought I have had is some kind of evolutionary tree, where one can see the evolution paths and pick any point and evolve from there. The tree would reveal what change caused the score jump, or more likely, upon review of the tree, what line contains the potential for the major score jump.

This type of UI presentation is a major undertaking but would also provide an excellent way of reviewing both soloist and group solutions as well as permitting user control of during a session to retry from any particular point an alternate approach. [This approach might address DisposableHeart's post.]

This would effectively replace the current linear undo, which looses information, (if you undo back 10 place and proceed, losing the 9 points), with a tree, where one could branch from any prior point, as many times as one desires. This would permit much more control and also provide a clear appreciation of which operations provided what gains.

The tree information along with the associated transscript of operations combined with a replay mechansims would permit playback review of any line, soloist or group. It awould also be an excellent research tool for ducumenting and understanding both automated and human folding solutions.


Annoatationsof the evolution....

One additional advantage of this type of record would be the ability to have the Folder annotate her/his thought process of the solution. Both the successful operations and the unsuccessful ones. A recorded talking protocol of a Folder's thought process recorded while watching the playback of the tree: "I started by doing
1. a shake - > wiggle
2. a wiggle -> shake
3. set some bands ->
.... a. shake ->
...... i. wiggle 30 ticks -> remove bands -> wiggle
...... ii. wiggle 10 ticks -> remove bands -> wiggle
...... iii. wiggle 5 ticks -> remove bands -> wiggle
...... iiii. wiggle 3 ticks -> remove bands -> wiggle
based on the score and the appearance I picked line 2. wiggle -> shake to proceed from there.


Additionally the set of operations could be condensed into a macro explantion, for example "3 by 2 Walk of the backbone" would be 100's of operations but conceptually be condensed into a single annotated transition on the evolution tree.

How'd she/he do that?

The tree along with the operation transcript would allow anyone who desired to do a forensic evaluation of the solution starting at the end and following the path back to the beginning.
Easy protein or efficient Folder?

One could also derive a metric of the path length start to finish vs the total tree length and derive an either an efficiency rating of the solver and a difficulty rating of the protein, or both.

Looking at a number of the best solution and the average path length and tree length could also provide a nice metric. This metric could alo be used to compare human versus automated solutions to compare efficientcies.


Now if there were only 100 grad students and 10 PhD canidates with copious free time to get crack'n on this wonderful problem domain....

End Blue Sky mode...

Joined: 05/19/2008
Groups: None

I don't have a solid alternative approach, only some ideas.

I'm only saying, that the current evolver scoring and ranking is not of the benefit of the FoldIt project. The reason I already mentioned in my opening post.

Solution? Well.... The scoring should not only take into account the highest score reached, but also the points increased. An evolver player should decide to fight for a few hard points on a top solution or make more easy points on a not-that-top one. This would encourage evolver players to work on other solo solutions submitted at a later stage.

Would evolutions overlap? Yes, and no. Any new evolved solution is a new branch of possibilities to others. Not only ONE solution is evolved into the top, but any (and all) other solves have the chance to be the best.

Would this imply that efforts are spread too wide and players go for the lower score increments and not for the top ones? Some most probably would. But if we consider how much effort is spent to achieve hardly no increase....

Would this help in player retention? Yes. There are rookie group members even in top groups that are happy to see how top solves look like, the more the better. And they contribute!

Would this help to work out as many solves as possible to the greates extent? Sure it does!

As for the changes in the app? I see only one change: the app has to keep track of the sequence of evolution on a per submited solution basis. (Any solve ever should have a recorded history of evolution.) The rest is some SQL queries to calculate global points after the puzzle is closed.

Another lovely idea: the original submitting player of the best ever solve should be honored with bonus points. :)

Joined: 05/26/2008
Groups: Hold My Beer

Another suggestion:

Perhaps it would be a good idea to further isolate solo and group play. Something I think would encourage both solo and group play, even by those not currently interested, would be:

Release a puzzle, and allow only soloist play on the puzzle. After the puzzle expires, pick the 3 best solutions, and release them as a new puzzle, group play only, and offer the 3 puzzle as the 3 starting points.

For example:
Mon: release puzzle 155, solo only
Fri: 155 expires, issue puzzle 156 as solo only
Sat: 155G1, 155G2, 155G3 are released for group play.
Wed: 156, 155G1, 155G2, 155G3 expire, release 157.
Thr: release 156G1, 156G2, 156G3...

I think this would not only encourage people to try different approaches in the early game since they are all competing solo, but would also give everyone an opportunity to work on a very good group solution, so they will not feel like they are hopelessly pitted against a team that is far ahead of them because one person got a good solution.

Joined: 06/17/2010
Status: Open » Closed

How team manage own evos is a team way of play.
Evolve scoring changed, closing.


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