Once upon a time (seven years ago), a Spork was talking to S.M. on the phone. “I just found a new flower, a beautiful purple one. Wait until you see it. You will be very happy.”
At that moment, three deer cruised through the yard, approached the aforementioned purple flowers, and ate every one of them.
“Never mind,” the Spork said. “Forget I ever mentioned it.”
The flowers were not seen again until this week. Perhaps some roots survived and lay dormant until this Spring. More likely, the deer have been eating this plant down to the soil level at the first sign of a seedling.
This year, fewer deer. More flowers.
“It’s a Campanula Glomerata (Superba)” — Surly Man who is happy with his photograph
Four coasters made from a Photoshopped photo of ice
(I used www.shutterfly.com to get these printed.)
Cardinal fledglings. Sometimes you just get lucky. Observing the yard is interesting, because it is unpredictable. Weather, changes in gardens, and even more sunlight because so many trees came down in Winter storms all contribute to an everchanging scene.
On the other hand, when we are working with technology, predictability is our goal.
Technical note: When photographing small objects with the optical zoom, it is better to use the remote rather than the shutter on the camera itself.
Spork has been critical of Web 2.0 services which try to predict users’ tastes based on analyses of other “similar” users, although she is also impressed with these recommendation systems when they work. The above book gives algorithms, theory, and practical programming advice on how to implement recommendations and other uses of large-scale data on one’s own website. The programming language used in the examples is Python, but the book explains the code so that a programmer unfamilar with Python could rewrite it in their favorite language or learn Python while reading the book. The book is fun to read and it gave me many ideas for possible projects.
Unfortunately, at least for this Spork, the difficulty of acquiring a suitable dataset for a real application and the gap between the toy examples shown in the book and the complexity of creating a web project that is actually useful and interesting has so far made it too daunting to attempt a project of my own. But the book nonetheless provides insight into how these “smart” websites work and makes it easier to understand how they go wrong.
The wide range in quality of recommendation systems among established websites illustrates the inherent difficulty. For example, Amazon’s book recommendations are quite excellent, but for other products their algorithms fail in almost comical ways: After I bought a hand-held mini-vacuum a few years ago, Amazon repeatedly suggested vacuum cleaners of all types. This is silly because once someone owns a vacuum cleaner, they no longer want to buy vacuum cleaners. Clearly, Amazon was using the book algorithm where it doesn’t apply. While a lover of postmodern poetry may want to acquire a large collection of poetry books by similar poets, few people want a vacuum cleaner collection.
Some other sites have truly awful recommendation systems: Audible.com’s is so bad (“People who like Anna Karenina also like the Da Vinci Code”) that it makes using the site more annoying. And last.fm, while it does an excellent job of finding music similar to other music, makes the user work far too hard and wait too long to get usable results, an unforgivable sin in the instant-gratifcation culture of the Internet. Then there are systems which make too many assumptions about our tastes, such as the famous “My TiVo thinks I’m gay” problem, in which TiVo supposedly recommends tons of gay-themed programs if a user chooses to watch a single show with a gay character.
Failed morning inspection after the Sporks went to bed before Surly Man.
Spork’s Swollen Eye
click image for larger view
Supposedly, this widget plays music I like. I think it does, but not very creatively, since it eventually runs out of songs once it runs out of the songs it has already played for me that I approved of. It is supposed to be able to extrapolate and suggest songs I might like based on who else likes the same music. It promises to do more “in a couple of weeks” which is odd since I thought computers were supposed to be fast…
[video]
[video]
