Saturday, July 26, 2008

GSoC Update

It's been a pretty laid back week or two for me. Google paid up, and I bought myself a microKorg (to compliment my circuit-bent Casio Sk-1), and I don't have buy my potatoes on credit anymore, which feels good, but a little un-American.

I spent most of these couple weeks just wandering the code, putting an end to bugs that stumbled across my path, but there are couple noteworthy items.

First, I have saving playlists more or less working. Which was something I was putting off writing, because it was a hassle, and not the kind of hassle I find very interesting.

The other good news is that the solver is about to become much much faster. I switched from hash table sets (QSet) to bit array sets, which are much better suited for the obscene number of intersect and subtract operations I have to do. The result is the solver should be nearly instantaneous. No more 20 second waits when you put in a bunch of biases.

I have a pretty expansive todo list, but I'm eager to implement some new types of biases to see what the solver can really do. I've gotten some good suggestions for biases that hadn't occurred to me. Let me know if you have an idea for one, and I may write it up.

Tuesday, July 15, 2008

Roll Video

I always loved in grade school when the teacher was too depressed or hung-over to teach the class and just put on a VHS of Raiders of the Lost Ark, or something.

As promised, I made a little demonstration for those not daring enough for the svn or nightly (you can admit it, we won't think less of you).

(thats: http://www.youtube.com/watch?v=CFg0313x-iU, for the embeded video challenged.)

Sorry about the pixelation, it's hard to make out how ridiculously hip my music collection is.

It's still a little slow for a lot a biases. Expecially on my aging thinkpad, while trying to record video. That's something I will be working on. The algorithm is gradually growing faster and more complicated. I suppose that's how that works sometimes.

For next week, I hope to get saving and loading ironed out, as well as putting more work into the solver. After that I have the much more interesting task of writing news types of biases and making sure they don't take forever to solve. I've already got a lot of interesting suggestions which hadn't occurred to me. It's going to be fun to see what this can really do.

Monday, July 7, 2008

Weekly Update: Unhidden Biases

Now that dynamic playlists are working, what's missing is a way to actually use them. That's what most of this weeks work went towards.

In Amarok 1, dynamic playlists are one of those great hidden features. They are terrifically powerful and usefull, but a lot casual users don't know about them, or just don't understand how to use them. I had been using Amarok for a while before I discovered them.

One of the problems I've been working on is how to present Amarok 2.0 dynamic playlists in a way in which it's more or less obvious what they do. I want it to be an exhibitionist feature, with the functionality all exposed and out in the open, no dialog boxes or menu hierarchies.

This week I spent some quality time working on the bias editor (and reminding myself why I don't like gui programming). Since every blog post needs a picture, here's the latest version of the bias editor from just a few minutes ago (not even in svn yet). It's a bit rough arround the edges, but comming along nicely. 


That screenshot is slightly faked, but in the next few days the bias editor will start to become functional. I'm pretty excited to see my work actually usable. 

I also did some important work behind the scenes. This sort of playlist generation can't really be done efficiently in general (it's an NP-hard problem). If you give it a really tough set of biases, it won't break down or freeze up, but it may give you playlist that isn't perfect. Making the solver work well is a matter of heuristics, trial and error, and tweaking. I did a little of that this week, but the real testing and tweaking will happen when the bias editor is up and running.

I'll be back here in a week, hopefully demonstrating how to use the new dynamic playlists.