One more thing, before I go to bed: An AmphetaOutlines update.

I'm not sure how many of you are still using the thing, but I've been using it hourly since I embarked upon the experiment. And then, very recently, the thing became insatiable with desire for my CPU and memory.

Turns out, in my spiffy new XML channel/item metadata files, I wasn't deleting data associated with old and no longer available items. This resulted in multi-megabyte XML files which AmphetaOutlines happily munched through for each channel to which I'm subscribed. Well, this update now regularly cleanses those files, leaving metadata stored only for those items that appeared in the current update of the channel.

So, if AmphetaOutlines has been becoming a dog for you, you might want to give this a shot. Upon the first run, the new code will wipe old data from the files. If your poor, battered machine can't survive another run in the current circumstances, then wipe the contents of data/channels_meta and start again. (But don't wipe your subscriptions or channel data! Just the channels_meta data.)

Let me know if this does good things for you. In the meantime, I'm thinking about what I could do by applying these ?BayesianAlgorithms (and those not-quite-so-BayesianAlgorithms) people have been tinkering with for use against spam. What if I could have AmphetaDesk initially sort my news items into ordered buckets of interest, according to my past viewing and scoring behavior? I really need to do some machine learning research.

Hell, what if I could go further and have a spider crawl blogrolls, looking for weblogs that seem to match other things I find interesting? Seems promising, though I think I'm still too naive about the subject.

Okay. Time for bed.


Archived Comments

  • I'm glad we have a very loyal beta-tester on this skin! :-)
  • Well, I am using this, so I installed the new version. No news, is good news :-)
  • Kick Ass! AmphetaOutlines is way cool. I use it all the time now. :-)
  • AmphetaOutlines and Mozilla's mouse gestures make Amphetadesk usable. My only beef is that I have to open an item in order to hide it, and some items do not have links. A "Hide?" checkbox beside every item and a "Hide Checked" submit button might help. In one pass, I could hide everything I don't want to bother opening (or can't open). In a second pass I could open all the remaining links. Something to think about.
  • Ooh, and how are you using Mozilla gestures with AmphetaDesk, might I ask? As for your suggestions... very good. I think I'd been considering that, but I really need to put a wish list together for the next iteration. I want to have more manual input into the thing - ie. manual item hiding, manual channel rating/karma
  • Another issue: Clicks are counted, and this determines placement on the page. But what about blogs like instapundit (at or scripingNews2 feeds, that don't have link tags? They can only migrate to the bottom. Currently, I fix this manually (I add the click tag to the meta data file).
  • Ehud: Yuck. Yeah, that's something I've been thinking about too. Not quite sure how to address it. At one point I thought maybe I could do a search-and-replace on href="" to wrap URLs with the click count redirect page. But that sounds like it could be messy. I was also thinking of counting outline expansions somehow add positive feedback toward a channel. In the next iteration of this, I want to give the channel a "score" that's based on clicks, manual karma rating, and maybe some other things. And I'd want to give each source of feedback a weight - ie. manual rating carries more importance than clicks perhaps. Hmm... off to make that wishlist page.
  • I thought about these ideas too. Exapnding an item, shows you care, doesn't it? :-) I think the bigger problem is that we need to make the list hierarchical (i.e., a tree). For example, there are thread that update regularly, and I read daily. But others, near the bottom of the page, update only once in a while, but when they do, I almost always want to read the full item. I am not sure a linear (1d) scale is what you need. Since I am a fan of machine learning, I can see how this becomes a multi-dimensional problem. Thinking from the user interface pov, I see a tree widget somewhere in the near future...
  • there are thread that update regularly ^^^^^^^ feeds Sorry