1/20 - 1/21

[RangerDenni]

I need to get over my fear of doing reviews because I have a lot to say and that's okay. I write many reviews but do not post them because they are very silly or strange or too long.
I also need to open up a little more around here and not be afraid of weirding people out or over-do-ing it.

I really like it here though and I have lots of ideas. I try not to jump up and down a lot but SongFighting takes up a lot of my time. In a good way.

[Caravan Ray]

CRay: The big problem with a simple +1/-1 is that it unfairly biases really high (or really low!) scores towards older content. The Elo rating scheme is much more fair, as newer and overlooked things can move much more quickly to its natural rank within the ecosystem (and part of its effect actually causes natural churn at the extreme edges, since if you have an established high-rank item that goes up against an unestablished or low-rank item, the low-rank item actually gets points no matter what - in the chess world, a newbie going up against Kasparov will give the newbie a lot of points no matter how it turns out, while Kasparov loses points just by merit of having gone up against the newbie anyway).

What I'd be afraid of with standard scoring systems (+1/-1, average rating, etc.) is you'd normally have a few very deeply-entrenched songs that stay at the top, without any breathing room for new things to show up. This is what happened with SomeSongs, for example. Elo has a built-in "freshness" factor that has nothing to do with arbitrary things (like "ratings in the last week" or whatever), and that combined with a random-pairing what's-better system means that you aren't just going to have people listening to the top few songs and going "yep, still like it" and keeping them cemented there.

Yeah. I do know a bit about how numbers and stuff works. I do that professionally.

But for fun - I like hearing 2 songs and picking a winner. That is the point

[Lunkhead]

Say I made a page where you could hear two random songs from the archives and pick a winner, then I made a page that showed recent winners. Would you guys really use this? Would you want to keep track of the winners you've picked, and/or see who picked the recent winners? If so you'd have to sign up/in to the Jukebox, so would you bother to do that?

[fluffy]

Could you just track all things with the Elo scoring system (as I linked to earlier) and then have a leaderboard? Being able to filter it by how recently they got their most recent winning vote would be nice, I guess, but with Elo it shouldn't matter as much.

Also if you track the number of wins and losses for each one, when someone picks a winner or loser you could apply simple Bayesian logic to say how often the voter agrees with everyone else, which might be fun.

I wouldn't personally be interested in tracking my winners but if it's easy to do, might as well. But only do that if people are signed in, of course.

(Also, the terminology on the site probably shouldn't be "winner" or "loser," just "liked more" and "liked less" or something. The actual internal scoring should also probably never be made visible.)

[ken]

I think the main thing is that we implement a way to "like" things and also to become a "fan" of things. This seems to be a very successful model on other popular sites.

[Lunkhead]

fluffy, I don't have time to do anything that complicated, I don't think. I'm just thinking that if there were a group of people ranking one random song over another, that could generate a stream of songs of interest that other people could use to possibly discover songs they like. So all I'm thinking is, one page, where you get two songs, pick a "winner", repeat, and another page, with a scroll of the "winners". (Substitute whatever your preferred term is for "winner".) I could probably bang something like that out pretty easily in the scant free time I get for programming.

[Lunkhead]

Oh, and that leads me to another idea for a way to rejuvenate the community: more software developer activity.

[fluffy]

Is there an API for the jukebox data store? That would make it easier to get more developer activity.

Also, the Elo rating scheme isn't that complicated - it's only slightly more math than the trivial +1/-1, although finding a nice clear simple score adjustment algorithm online is a lot harder than I thought it would be. I'm just thinking that being able to have some sort of scoring mechanism and being able to look at things based on what's scored the best is the fun part of this. It doesn't have to be Elo, of course. It just has some nice mathematical properties that work well for this use case. :)

[Billy's Little Trip]

Oh, and that leads me to another idea for a way to rejuvenate the community: more software developer activity.

I'd help if I could. But I'm more of a hardware developer. HEY OH!

[Lunkhead]

fluffy, you can append ".json" or ".xml" to any Jukebox fight(s)/artist(s)/song(s) URL to get the data in those formats. (".m3u" works also though that's less useful for API purposes.)

http://sfjukebox.org/fights/our_plans_are_on_hold.json

MG's iOS App is using the JSON URLs I believe.

Providing easy programmatic access to the archive data was part of my original goals for building the Jukebox.

[fluffy]

Ah, and /fights.json gets the list of all fights with entry counts. Perfect. Could that also be extended to give the number of songs in the archive? (I know I could compute it myself, but you already have the number of entry rows in the database available, presumably.)

Presumably you'd prefer if anyone who queries /fights to cache it with a fairly large TTL (looks like the query takes a little while, in any case), but that's straightforward and it's not like the data changes more than once a week.

[fluffy]

I'm also thinking of how to build something Pandora-like as well, while I'm in this headspace. Unstructured clustering used to be something I played around with a lot, and jeeze, I have ideas now (which also feed back into an automatic playlist generation idea I used to obsess about).

Totally going to do this thing. With all my copious spare time. Isn't that always the problem?

[Mike Lamb]

I think the main thing is that we implement a way to "like" things and also to become a "fan" of things. This seems to be a very successful model on other popular sites.

Mike Lamb likes this.

Although both of our posts are likely in jest, I think this is more accurate than not.. especially with "new blood." Forums in general seem to be losing steam in favor of FB, twitter, etc.; more "real time" communications? I dunno. Don't have any answer or suggestion, just an observation.

[Ross]

As interesting as the archive is, I don't think that getting people to listen to it is the main issue. i think that encouraging ongoing conversation of current fights is much more central to the flavor of the site.

I am not sure of the reasons for the current pacing of the cycle of songs, but I think the slower pace also slows down the feel of the boards. When there was seven day cycles, you had to say everything you were going to say in seven days, so the info came faster and people had to reply faster. i think that caused a feel of a more active site (and may have combatted procrastination).

[fluffy]

I think this site has many different flavors. I like to think about the reason Narbotic first started it: he was doing songs-by-title-suggestions, found it to be very lonely work, and decided to open it up to others to see what other people did with those suggestions. To me that's about taking a germ of an idea and exploring it in many different ways as possible.

I think trying different things and seeing what sticks is very much what this site is about, and keeping things going in one direction just because that's how things have always been is not a good plan. Same is boring.

If there's a part you don't like, you can always ignore it. Unless that's the only part.

[Lunkhead]

Ah, and /fights.json gets the list of all fights with entry counts. Perfect. Could that also be extended to give the number of songs in the archive? (I know I could compute it myself, but you already have the number of entry rows in the database available, presumably.)

Presumably you'd prefer if anyone who queries /fights to cache it with a fairly large TTL (looks like the query takes a little while, in any case), but that's straightforward and it's not like the data changes more than once a week.

I don't have a straightforward way to access the number of songs, or fights or artists, for that matter. I do make that available in the HTML for the homepage, though, so it would be pretty straightforward for me to add a "stats" endpoint that would return those numbers if you'd like.

Another thing is, the fights/artists/songs URLs can also take search parameters, as per the forms on the fights/artists/songs pages, and also "perPage" and "page" (1-based) parameters for pagination. One problem currently is that there is some hokey stuff going on with remembering your last search in your session, but it's a little flaky how it works. I need to sort that out at some point.

It's OK if you want to query /fights.json to get all the fights, I don't think it's going to matter as far as my hosting situation goes, unless you plan to hit that very frequently. It's pretty slow, though, so you might want to save that data locally if you want. I don't have JSONP support yet, but maybe I should add that in, to make the data available more easily on the client side.

[fluffy]

Yeah my plan right now is to just use the .json to pull the content into my own database instance and just do things the old-fashioned way. If I get around to it. I'll probably just use Carbonado and JSPs or something, if that's something I can set up easily enough on Dreamhost or my Linode or whatever. (I've been using Carbonado at work. It's pretty nice, and actually makes a great case for Java as a web programming language.)

[Lunkhead]

If you just want your own copy of the whole database, I'm not sure pulling it from the .json is a great way to get that. Do you want a mysqldump so you have a snapshot to play around with? Or I could send you my code that parses and imports the archive text file if you want. It's a bit hacky, as you might imagine, so another pair of eyes on it could be helpful. I actually could put up all the source to the Jukebox if anybody would care to use it. Any "secret" info about the inner workings of songfight.org has been abstracted out into config files that are not in the source, so it should be "safe" to post the code.

I'd never heard of Coronado, it sounds interesting. Does it support DBs other than SQL/BDB? I'm still using Spring and Hibernate, and finding them to be pretty useful for myself, especially Spring. Hibernate's not been under much significant development for a while but the Spring folks keep cranking out useful libraries (for social, mobile, NoSQL, etc.).

[fluffy]

Carbonado is designed such that you can generate bindings to pretty much any persistence library. Anything else I say would probably violate my NDA.

Spring is pretty great, but doesn't it also require a servelet hosting engine of some sort? I can't find any information about running Spring on Dreamhost, in any case.

[Lunkhead]

You can't run Spring MVC w/o servlets, I don't think. I'm not sure how you'd use Spring at all if you're only using JSPs, as you don't really have any kind of persistent context (e.g. a webapp in a servlet container)? I haven't tried to do things that way, I've always used servlets. (That's pretty "old school", like 1990s old practically.) Last time I checked, which was years ago, but a quick search turns up the same results, was that Dreamhost didn't support JSP/Servlet except if you set it all up yourself on a virtual private server?

http://wiki.dreamhost.com/Java

[fluffy]

Yeah, that's kind of what I figured. Oh well. I do have a VPS with Linode that I could use instead. Also this is something I can finally do with the wanderlute.com domain I registered a while back (I had some ideas for collaborative/exploratory music that I never had time to work on, and so here's another one).

[pegor]

Suggestions for amping up fight participitation and maybe even community involvment...

I know I frequently have a chorus or a guitar line for a fight but never follow threw and finish a song. I am to crappy/insecure of a musician to ask someone to collab. I'd like to see a 'Craigs list sorta place where I could browse for a chord progression, singer, etc, and also post my bits as free to a good home. It should probably be wiped clean for each fight to avoid staleness.

Another thought if there was a scrolling update of what's being said in IRC/twatter/FB on the main page or the forum page - it might spark more people jumping into the various discussions.

my coding skills are out of date, so uhh just ideas.....