Developer Chat

This was an impromptu feedback discussion on the new overhauled Remix tool. It was posted by popular demand.

jflat06 hmm well if the puzzle wasn't already on that client, then your saved solutions wouldn't be either. did you play that puzzle on a different client previously? 11:29
brunokestemont I think that I played it also on this devprev client, I'll verify on another computer. btw, the new remix tool seems interesting (the buttons appear only when opening rama map) 11:34
jflat06 which buttons only appear? 11:36
brunokestemont stop cross with arrows to navigate through remix proposals (when removing rama, it stays) 11:39
jflat06 interesting 11:40
jflat06 i tested it on all platforms and it came up when i pressed the remix button for me 11:40
jflat06 are you in the selection interface or normal? 11:40
brunokestemont normal on a mac 11:40
brunokestemont simple remix gives a stop button on left up, but we don't see anything more unless opening rama 11:41
jflat06 i'll give it a try 11:41
brunokestemont if I do anything else after running remix, the buttons appear (like even typing something here) 11:44
brunokestemont or clicking on the window 11:44
jflat06 ahhh, i think i understand 11:45
jflat06 for me it just takes like 10-15 seconds to pop up the first time 11:46
jflat06 if you wait, does it pop up eventually? 11:46
brunokestemont i'll try 11:46
jflat06 also, what puzzle are you attempting to use the tool on? 11:46
brunokestemont strange, now it pops up after 1-2 seconds 11:47
jflat06 yeah 11:47
jflat06 if you restart the puzzle, you will probably run into the long delay 11:47
jflat06 errr 11:47
jflat06 restart the client 11:47
jflat06 the delay is because the fragment library that the remix tool depends on isn't loaded until it's needed 11:47
brunokestemont btw I succedd to save client, close 1221 and re-open it. great ! 11:48
jflat06 cool! 11:48
jflat06 anyway, the client should only have the long-delay the first time you run remix 11:48
jflat06 and only if you aren't running a design puzzle 11:48
brunokestemont ok may be adding something like "loading remix library" 11:49
jflat06 (the client will auto-load the library during the loading screen for design puzzles) 11:49
jflat06 yeah, i'll probably just move it to the puzzle loading screen like it is for design pzuzles 11:50
jflat06 it was an oversight on my part that it wasn't pre-loaded for prediction puzzles 11:50
jflat06 hopefully the new version of the tool is much more useful 11:50
jflat06 i got the feeling that not many people used the old one 11:50
brunokestemont yes I like it, it's even interesting to look at it together with rama map 11:51
Brow42IRC I still don't really understnd what remix does, it usually says it can't find anything, and then it messes up my global structure when it works 11:56
brunokestemont Not beeing obliged to cut before is also better (hopefully, remix within scripts will also work better) 11:56
brunokestemont @Brow, new remix doesn't mess the global structure any more 11:59
jflat06 @Brow, you can also change the region it is operating on. sometimes selecting one more or one less residue on either side will get it to give good results 12:10
jflat06 also, i think it tends to give better results if you select a larger region 12:11
jflat06 for example, if you are trying to figure out a loop between two sheets, i select part of each of the stands, too 12:11
jflat06 like ~7 residues 12:11
Brow42IRC well I also use orginal interface 12:11
jflat06 you can still do it in the original interface, but yeah, that involves changing secondary structure to get it to do what you want 12:12
brunokestemont yeah ! succeded to load my old solution 1221 (even if the score is now -999999.999) 12:12
jflat06 if you want to use it with the original interface, someone will probably make a script that will try various lengths 12:14
jflat06 although at the moment, the new version of remix isn't supported in scripts 12:14
brunokestemont ahh good to know, I'll wait for scripting 12:14
jflat06 in my experience from testing this on players solutions though, it usually finds something anyway 12:16
jflat06 definitely encourage everyone to use it, since rebuild is really really terrible scientifically 12:17
brunokestemont what will the old script remix tool do? 12:17
jflat06 it will just do what it did before, basically 12:18
brunokestemont ok 12:18
jflat06 i haven't implemented scripting for the new one because right now there's no way of returning a bunch of results in Lua, to my knowledge 12:18
brunokestemont and something like one result per iteration? 12:20
@Susume2 you could pass in a list of quicksave slot numbers and put the results into those quicksave slots 12:20
@TimovdL You might consider something like I did in my stacking position, so put the results in slot n until n_m 12:20
jflat06 yeah, something like those might work 12:20
@TimovdL Great minds have same idea's 12:20
jflat06 i'm wondering if there's a built in data structure that can handle this, though 12:20
@Susume2 ;-) 12:20
@TimovdL Lua is untyped, you can always make an array of things 12:21
jflat06 you can, but as it stands, the tools all assume that the 'work' is done to the main game pose 12:22
jflat06 whereas the natural way of doing this would be to return a list of results 12:22
jflat06 right now, they all just return void 12:22
@Susume2 if lua had access to the undo graph you could lay down the results in the undo graph and leave it up to the script writer to copy them to quicksave slots or pick one to work from or whatever 12:23
jflat06 yeah, that's possible 12:24
jflat06 kind of what rebuild does right now, effectively 12:24
brunokestemont exactly 12:24
@Susume2 have the tool return the number of undo positions it populated 12:25
@TimovdL Or a function to find the nr of remix position and a way to iterate through them 12:26
brunokestemont crashed a new devprev ;( 12:27
@TimovdL Really love to make a DRemixW 12:28
@TimovdL Would be pretty easy 12:28
@Susume2 if remix returns better results than rebuild a DRemixW could overtake DRW as the most used script - that should make the scientists happy 12:29
jflat06 if you implement it, my biggest suggestion would by to try and use larger fragment sizes 12:29
jflat06 our biggest remaining issue with foldit designs is fragment quality 12:30
brunokestemont strange behaviour of 1221: infinite negative score, infinite shake iteration 1 12:30
@TimovdL Good to have this chat about it 12:30
jflat06 and i think that larger fragments will preserve the backbone best 12:30
@TimovdL But the larger more degrees of freedom 12:31
brunokestemont what do you mean with larger fragments? 12:31
@TimovdL So harder to find good ones 12:31
jflat06 as in selections of size 6-9 instead of 3 12:31
brunokestemont 6-9 seg for each loop? 12:32
jflat06 well at the same time, you'll get a lot of results back now 12:32
@Susume2 DRW currently does only one iter of rebuild, evaluates the result, then does another single iteration, usually getting a different result - if you give lua access to new remix returning only a single result the DRemixW script could be built 12:32
jflat06 @Susume2, that works because rebuild is random 12:32
jflat06 i'd like to keep it non-random for now 12:32
brunokestemont why non-random? 12:33
@TimovdL Then I think the easiest would be to give a slot list as input and return the nr of slots filled 12:33
@Susume2 longer fragments with remix hopefully will deform the protein less than longer fragments with rebuild 12:34
jflat06 several reasons for non-random 12:34
jflat06 biggest one is that it makes testing and debugging 1000x easier 12:34
brunokestemont ok 12:34
jflat06 it's partly a lack of any real reason why it SHOULD be random 12:34
jflat06 @Susume2, yeah 12:35
@TimovdL what criteria does it use to pick one over another if any? 12:35
jflat06 if you run the new remix on small fragments (like size 3), instead of deforming the protein now, it will just get worse ideality scores 12:35
jflat06 longer fragments should have an easier time closing the cut, and will also have the benefit of being a longer stretch of high-probability backbone 12:36
jflat06 @timo right now the criteria is just to go through all the available backbones and close them 12:37
brunokestemont argh, a pity that the remix arrows appear just over the loop 12:38
jflat06 i think internally it does some small checks based on the sequence to make sure the backbone being returned isn't violating something in the torsion space 12:38
jflat06 @bruno, i can move them down a little bit 12:38
jflat06 but the list of fragments is essentially unordered, i think 12:39
brunokestemont soltion nb 1 is the current one, shall I prefer lower scoring solutions 2_n? 12:40
jflat06 sorry, what do you mean? 12:41
brunokestemont I try remix on a loop and I get 3 proposals, the first one beeing what i have 12:42
brunokestemont nb 2 and 3 score lower 12:42
jflat06 solution 1 will always be whatever was there before - i left it like that so that you could compare more easily 12:43
jflat06 so if you accept the first position, it isn't actually doing anything 12:43
jflat06 as for deciding on which one to pick 12:44
jflat06 i think a lot of that is just visually looking at them and deciding which one is closest to what you want 12:44
brunokestemont the difference with rebuild is that I might expect that nb2 and 3 are always scientifically better? 12:45
jflat06 just like rebuild, the score isn't terribly meaningful since the results have not been shaken 12:45
@Susume2 one very common cause of scoring lower will be sidechain clashing, so in a script, shaking each one before comparing would be smart (as DRW does) 12:45
brunokestemont ah ok 12:45
jflat06 @Susume2 exactly 12:45
jflat06 I may try and include the centroid score of the results in the UI as well, so that you can get a feel for the score without having to shake them 12:46
@TimovdL centroid score? 12:47
jflat06 yeah, internally tools like rebuild use a scoring that just thinks of sidechains as sort of blobs so that it doesn't have to shake things 12:48
jflat06 thats how rebuild picks which results to show you 12:48
@TimovdL That is why rebuild never gives backbone clashing f.i? 12:49
jflat06 f.i? 12:50
@TimovdL for instance 12:50
jflat06 yeah 12:50
jflat06 centroid scoring still considers the full backbone 12:51
jflat06 sidechains are just approximated 12:51
brunokestemont problem if I use the tool, I've to stop and shake then remix again to compare 12:51
jflat06 we could shake the results, but that would probably slow it down significantly 12:52
jflat06 i guess my hope is that visual inspection of the results will help you decide 12:53
@TimovdL bruno, can';t you use the undo's to go back in the list of results? 12:53
jflat06 @Timo, no, I made it only push an undo at the end 12:54
jflat06 I think Undo's should be a single player decision/action 12:54
jflat06 but I think there might be ways of accomplishing the same thing 12:55

Generated by irclog2html.py 2.13.1 by Marius Gedminas
- find it at mg.pov.lt!

Get Started: Download
  Windows    OSX    Linux  
Windows
(7/8/10)
OSX
(10.7 or later)
Linux
(64-bit)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Search
Only search fold.it
Recommend Foldit
User login
Soloists
Evolvers
Groups
Topics
Top New Users
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