2016.04.01 Dev Chat

From Eterna Wiki

[13:59:50] <ChatBot> KA-BOOOM!!!

[14:02:55] <whbob> Hi

[14:03:21] <bekeep> Hi whbob

[14:03:39] <bekeep> I don't think Jen can make it today.

[14:04:25] <whbob> I haven't attended many chats lately.

[14:04:46] <bekeep> Me neither. : )

[14:05:39] <whbob> I'm getting a mountain of data on labs I started last August :)

[14:06:13] <bekeep> Nice!

[14:07:00] <bekeep> Have you found anything interesting?

[14:07:16] <bekeep> What's your approach going to be?

[14:07:57] <whbob> My lab solutions cover the full range of scores with more closer to zero than 100 :)

[14:08:51] <whbob> With the latest lab, I made mod's of the highest past lab to see how I can distort them.

[14:11:17] <whbob> I'm curious to know if anyone has explained what A*B/C^2 would look like in a lab. Three oligo's?

[14:11:37] <bekeep> 4 oligos and 4 states

[14:11:46] <whbob> Wow!

[14:11:47] <bekeep> Nando has an example puzzle on the dev side

[14:12:07] <bekeep> The problem is, it takes a long time to compute the folding

[14:12:14] <bekeep> So Nando created a new "freeze" button

[14:12:24] <bekeep> that will freeze the folding calculations until you unfreeze them again

[14:12:33] <bekeep> And then it will take a few minutes to compute again

[14:12:48] <bekeep> The 4-state version has 1 oligo A, 1 oligo B and 2 oligo Cs

[14:14:04] <whbob> With each oligo, there's another order of magnetude of computation? I'm gonna need a bigger computer :)

[14:16:04] <whbob> I haven't been able to come to any conclusions about the RNA oligo's to use for real TB testing.

[14:16:23] <bekeep> Yeah... it's kind of crazy... I'm not sure if there's ways of simplifying the problem

[14:16:38] <whbob> Real mRNA's from blood or processed mRNA's.

[14:16:39] <bekeep> Ah - the forum post that Omei put up?

[14:17:34] <whbob> Read it, sounds like it's more of a tech question as to whats practical in the field.

[14:18:56] <whbob> Do we have many new comers? Some are progressing very well I see.

[14:19:13] <bekeep> Yeah, I haven't been following it at all, but I find the question very interesting... I like that the answer lies at the intersection of lab process expertise, knowledge about how the ultimate device would work, and the expertise of design from the players

[14:19:29] <bekeep> Was just chatting about retention with the dev team

[14:20:07] <bekeep> The player written paper gave us a burst of new players, but I think some technical problems (next puzzle button not showing up) and some poorly designed leveling has discouraged some of them

[14:20:50] <bekeep> Still, it's nice to see a big chunk of folks progressing. I think we have ~20 new players who have made it all the way through the progression (which is saying something, given how hard it is)

[14:21:20] <bekeep> There's a lot I have yet to look into, but I've identified problem puzzles from dropout rates and # solves / # of attempts

[14:21:22] <whbob> Yes, It's more challenging than when I started.

[14:21:57] <bekeep> And hoglahoo, who has really been the person to make it all work by helping players through trouble spots, has also given me some advice

[14:22:16] <bekeep> There should be a new iteration (with a couple of added puzzles, plus some other adjustments) coming in the next week or so

[14:22:32] <Astromon> sounds good

[14:22:59] <bekeep> The thing that I'm happy about is that we're tracking player success now in a way we never did in the past, which gives us the opportunity to improve

[14:23:03] <bekeep> Hi Astro

[14:23:05] <Astromon> hi beekeep and whotbob

[14:23:19] <whbob> I still have about 7 switch challenges that I can't solve. Mental blocks on all of them:)

[14:23:22] <Astromon> was reading what i missed

[14:23:26] <whbob> Hi Astro.

[14:23:50] <bekeep> haha : ) There's a couple of Super Computer ones that are pretty freaking difficult for me

[14:24:07] <Astromon> looking forward to the a/b/c labs&lt;&lt;

[14:24:31] <bekeep> If you want to send me your thoughts (or even just the puzzle ids and your attempted solution) it might be helpful in me seeing where people can get stuck

[14:24:37] <Astromon> they are hard but i think it must be so to train for labs

[14:24:59] <whbob> a/b/c makes my head hurt :)

[14:25:08] <bekeep> There are some puzzles that have high variances: I can play it once and find the right path but if I return a few weeks later I end up in a dead end

[14:25:18] <whbob> @ bekeep, will do.

[14:25:19] <Astromon> im sure it will make my head hurt also haha

[14:25:25] <bekeep> Agreed.

[14:25:55] <bekeep> Yes, the Super Computer ones are really hard, but good training.

[14:26:48] <whbob> I think there is a link now that gets you back to the comments page for hints.

[14:26:56] <Astromon> new players though have passed them

[14:27:06] <Astromon> yeaH

[14:27:17] <Astromon> with a little help haha

[14:27:22] <bekeep> Yeah, it's not exactly ideal, but it's there. Hopefully that's helping. And Caleb fixed the next puzzle button to always float above the others.

[14:27:41] <Astromon> nice that was a mojor prob

[14:27:50] <bekeep> Yeah... took too long to fix, but glad we did

[14:28:46] <bekeep> I moved structure energies to be introduced slightly earlier, so that it can motivate boosting, and added some example boosts in the earlier puzzles (that are locked) to prime the idea a bit

[14:28:51] <Astromon> 4 states what April fools?

[14:28:54] <bekeep> Also added some locks/simplifications

[14:29:01] <bekeep> Nando's a joker

[14:29:05] <Astromon> is it nupaK?

[14:29:15] <bekeep> Yes, it's nupack

[14:29:23] <Astromon> oH

[14:29:47] <whbob> It would be interesting to hear from players who give up.

[14:30:05] <bekeep> I think it'll be released next week or the week after... wasn't paying attention on that point in the dev meeting

[14:30:09] <Astromon> so i can relate to solving some puzzles then not being able to do so

[14:30:12] <bekeep> @whbob - YES

[14:30:24] <Astromon> the current R3 puzzle i solved last round

[14:30:45] <Astromon> i was wanting to do a new design for it and am unable to do so

[14:30:48] <bekeep> One thing I'm interested in is getting move sets from the players who didn't solve the puzzle, but I think right now, as the flash is set up, there's not a convenient way of sending that data to the servers

[14:31:03] <Astromon> hmm

[14:31:24] <bekeep> Yeah... there's a certain path dependency to the problem solving process

[14:31:44] <Astromon> i guess i solved it on accident haha

[14:31:55] <whbob> I have been looing at chess programs and how they save player moves in a game. Gives the history of the game.

[14:31:58] <bekeep> When we walk down a path I think we end up making different assumptions about what should be fixed and not...

[14:32:18] <Astromon> my best score was a 50 though i moded it in the current labs try to improve it&lt;&lt;

[14:32:19] <bekeep> We save player moves for submitted solutions

[14:32:43] <Astromon> oH wow

[14:33:07] <Astromon> the moves are saved then i was wanting to see how players designed certain puzzles

[14:33:08] <bekeep> (it's in the EULA, actually... I know not a lot of people read those.)

[14:33:13] <whbob> Wondering jow we might be ablle to take advantage of that method.

[14:34:02] <whbob> *how

[14:34:18] <bekeep> So I think from Rhiju and the gang's perspective it can be helpful in refining Eternabot (or creating other bots) by creating a policy network of "expert moves" similar to that used for Alphago, the program that beat one of the top go players 4/5 games

[14:34:46] <Astromon> thats interesting

[14:35:07] <bekeep> I'm interested in seeing what moves players are attracted to as a proxy for the concept/approach they might be employing

[14:35:22] <Astromon> indeed

[14:35:31] <Astromon> that would be great Data

[14:35:33] <bekeep> And looking at the diversity of player moves in relationship to the diversity of solutions, particular for labs

[14:35:46] <andrewhaeffner> I was so amazed when I read about Alphago

[14:36:46] <bekeep> I feel like the move set data should be available to everyone as well, but I don't think it is currently. That was the original concept. Not sure on whether it will be incorporated into the latest version of the data browser (I certainly haven't heard discussions along those lines)

[14:37:38] <bekeep> Yeah, it's an interesting achievement. Also interesting to see the reaction in the press. I feel like sometimes the media misrepresents how AI works.

[14:38:08] <bekeep> Lee Sedol had a tough time - playing against lots of accumulated wisdom of many other go players... (avid go player here : ) )

[14:39:37] <whbob> Maybe the data browser could link a design ID# to a "move" file that has the record of moves to solve it?

[14:40:19] <Jumanji> how!?!?!

[14:40:25] <bekeep> That would make sense

[14:40:36] <bekeep> @Jumanji - not sure what you mean

[14:41:10] <Jumanji> Im on gene sythesiser 10 and just cant figure it out.

[14:41:40] <bekeep> Ha - you came to the right place. : )

[14:41:46] <bekeep> Have you taken a look at the hint at all?

[14:42:19] <Jumanji> yes, ive gotten that far, it's just this one loop that's giving me trouble

[14:43:14] <bekeep> Which loop - the center one? (the "multi-loop")?

[14:44:05] <bekeep> (and sorry that you've had so much trouble)

[14:44:14] <Jumanji> I got the center one, but it's the branch that comes out the bottom of the center

[14:45:01] <bekeep> Hmm.. what's the free energy of the center one?

[14:45:10] <bekeep> (feel free to also share screen shot if you'd like)

[14:45:46] <Jumanji> http://www.eternagame.org/sites/default/files/chat_screens/228675_1459539938.png

[14:46:16] <Jumanji> sorry, it was in natural

[14:46:31] <Jumanji> http://www.eternagame.org/sites/default/files/chat_screens/228675_1459539987.png

[14:47:18] <bekeep> Okay, looking at target mode on the lower branch, notice how the base pairs aren't paired up?

[14:47:27] <bekeep> A G and an A won't form a base pair together in a stack

[14:47:28] <whbob> 49 is a mis match.

[14:47:43] <bekeep> Nor will a C and an A

[14:49:09] <bekeep> Base 41 and base 52 needs to be an AU pair

[14:49:24] <bekeep> Strengthen all of the bonds in that lower part to be the strongest possible

[14:52:27] <Astromon> 37 has to be a U

[14:53:06] <bekeep> I would also go to game options (three bars on the left-hand of the control panel) -&gt; settings -&gt; turn -&gt; display free energies for all structures

[14:53:06] <whbob> @jumanji; the base pairs in the multiloop "might" line up in one direction better than another.

[14:53:35] <bekeep> The free energy for the structures will help guide the puzzle solving process

[14:54:09] <Jumanji> Thank you so much guys! I got it!

[14:54:16] <bekeep> Nice!

[14:54:26] <Astromon> congrats!

[14:54:32] <whbob> Congrats!!

[14:55:52] <whbob> 12 minutes to solve. Not bad Jumanji! :)

[14:57:56] <whbob> @ bekeep: Looks like you and the other Dev's are working at warp speed. Doing great!

[15:01:49] <whbob> @ Astro: R3 - I couldn't see a solution that didn't require having to take advantage to the specific oligo's in the puzzle. Tough to solve.

[15:01:53] <Astromon> i thought it would be easy for the computer to save ones move in a puzzle i guess its not so simple

[15:02:41] <Astromon> yeaH i solved it pretty quick the first time around somehow

[15:03:18] <Astromon> also its hard I think because there is no guidance on which oligos need to be bounded

[15:03:21] <whbob> @ Astro: it's just saving a ascii file of sequence number and base type for each "move", not too hard:)

[15:03:32] <bekeep> Thanks whbob!

[15:04:01] <bekeep> Yeah - whbob is right. It would just be a json file, basically

[15:04:12] <Astromon> cool lets do it!

[15:04:16] <bekeep> Although we've talked about ways of visualizing the move set

[15:04:46] <bekeep> But that would be more complex. Just talked to Caleb about creating something like that and he doesn't think it would be too hard.

[15:05:00] <bekeep> (creating the json file and linking to it I mean)

[15:05:01] <Astromon> we should be able to open a design then press the back botton to get to the beginning to see how it was solved!

[15:05:08] <bekeep> That would be awesome

[15:05:30] <whbob> I would want to have a manul and an auto step through with the ability to stop, go and reverse:)

[15:05:40] <whbob> *manual

[15:05:44] <bekeep> Yep - definitely

[15:06:52] <Astromon> i was thinking about making a tutorial on a lab sometime

[15:07:09] <Astromon> kind of lightshot my step by step moves

[15:07:41] <bekeep> If someone has programming experience they could take the json and do a step through (although, wouldn't be able to have fine-tuned control like whbob suggests)

[15:07:49] <bekeep> @Astro I think that would be fantastic

[15:07:50] <Astromon> would require alot of time depending on how many moves it takes me to pass all constraints

[15:08:07] <whbob> Move history: I'd be able to see how often I repeat the same wrong paths over and over again (painfully) :)

[15:08:19] <Astromon> haha

[15:08:20] <bekeep> We're trying to feature more links to lab tutorials/advice on the lab pages (rather than making people find them wherever)...

[15:08:27] <bekeep> No doubt - me too!

[15:08:43] <Astromon> oH thats a great idea bekeep yes!

[15:08:50] <bekeep> Eli's idea

[15:08:52] <bekeep> : )

[15:09:03] <bekeep> But we will try to get it working

[15:09:30] <bekeep> Right now we are just putting in links manually, so if you have a lab tutorial that you think would apply to the current lab, send it to Caleb or me and we can put the link up

[15:09:43] <Astromon> all good things takes time and patients :D

[15:10:12] <bekeep> Ideally we'd like a section of the lab to be for links to lab tutorials or advice for newcomers to the lab (or veterans!) beyond just the comments

[15:10:30] <Astromon> nice

[15:10:39] <whbob> Great.

[15:11:15] <whbob> Malcolm made a comment about a switch puzzle. Can't remember which one now.

[15:11:56] <whbob> It described that the bound state would have energy value + 4.86 or what ever molecule energy.

[15:12:45] <bekeep> This is a switch puzzle in the progression?

[15:12:46] <whbob> Then he added that the unbound state needed to be no more than the bound energy state with out the molecule energy.

[15:12:47] <Astromon> thats a nice specific giudence

[15:13:16] <whbob> No, it was just a reular challenge puzzle.

[15:13:17] <Trevor12> hi

[15:13:19] <bekeep> Ah

[15:13:26] <whbob> *regular

[15:13:28] <Astromon> hi trever

[15:13:35] <bekeep> hi Trevor

[15:13:43] <whbob> Hi Trever

[15:13:46] <Trevor12> im new

[15:13:55] <Astromon> Welcome to Eterna!

[15:14:04] <Trevor12> thx

[15:14:05] <bekeep> I have to go afk now, but feel free to ping me with more advice. : )

[15:14:20] <Astromon> nice dev chat bekeep thanks

[15:14:31] <whbob> Thanks Bekeep :)

[15:14:37] <bekeep> Thanks guys