Sep
29
2007
-
Site for the new “High Performance Web Site Techniques” book, plus an interesting sprite generator
-
A project analysing biographical materials, making some use of RDF too
-
Cool tool to convert unicode characters
Sep
26
2007
-
The libxml Python API is very lightly documented, so this is an attempt to fill in some of the holes that exist.
-
Contains an excellent tutorial on the Rete algorithm, plus significant enhancements
Sep
25
2007
-
A nice exposition of eventual consistency. Makes the point that there’s no difference to the end user whether a transaction failed because a forklift crushed the product or a dirty read said it was available when it wasn’t.
-
OAuth is a nice complement to OpenID, allowing limite access to machine APIs to be delegated. Good for connecting services together.
-
Wide tables and sparse data… isn’t that a definition of RDF
-
Vertical partitioning implemented with a column store
-
Describes the H-Store architecture (a step on from C-Store). Looks like some very interesting research is going on at MIT CSAIL
-
Michael Stonebraker’s company and column-store product
Sep
24
2007
Via ComputerWorld comes more news of how the genealogy world is hotting up. For example, FamilyInHistory.com provides:
Gone are the days when families filled neighborhoods and Aunt Geraldine lived down the street with your 7 cousins. Now you can keep in touch with a family blog. The blog is a fantastic way to share good news with the whole family, like a promotion or a new baby. Reminisce about childhood memories, share genealogy breakthroughs, get to know your distant cousins, or plan a family reunion.
And SaaS too:
If you are going on a genealogy scouting trip, you don’t have to pack up all of your loose papers. Just bring a laptop and access your family tree website online. If you travel for work, evenings can be hard to fill, but it’s a great time to work on your family history. You can also work on it during your lunch hour, waiting for a flight, or anywhere you have access to the internet.
I still don’t see anything that would attract professional genealogists who would favour an evidence based approach to research, but this space is certainly getting a lot of interest.
Sep
24
2007
-
How come I’ve never seen this before? He gives three rules, but only one is important IMHO: “Only solutions that produce partial results when partially implemented can succeed.” I think that’s true for other systems too c.f. agile development
-
Details of the 7.04% solution to the Netflix Prize. Regularised SVD with biases, postprocessed with kernel ridge regression
-
RBMs slightly outperform carefully tuned SVD models
-
Long post with lots of gruesome but readable details of SVD approach to the Netflix data
-
A Bayesian approach to counter overfitting of SVD caused by data sparsity (as in the Netflix data)
-
Uses Principal Component Analysis extended with variational Bayesian learning
-
Expression of Bayesian concepts in an extension to OWL
-
A vicabulary for Bayesian concepts
-
More on representation of Bayesian networks in RDF with a comparison between it and BayesOWL
-
Kendall Clark: “I anticipate that, at some point, we will talk about, say, an RORB (RDF, OWL, Rules, Bayes) platform for Semantic Web development.” IMHO OWL isn’t part of the petatriple future of the semweb. Nor is SPARQL…
-
Interesting points, although I think he missed the biggest reason: don’t rewrite. It always costs more than you think, especially when you have years of obscure bugfixes and workarounds that you have to reimplement.
Sep
24
2007
I dug up some links on the mismatch between MVC and the Web:
It’s hard to explain in simple terms why MVC is such a poor pattern for web development. I don’t claim any particular success in converting people away from the lure of GUI patterns but in my experience the hardest task is resetting people’s internal model of a web “application”. MVC works pretty well in the traditional GUI where you might have an item (e.g. a document) that you wrap with a model and expose with different views (as an icon and label in a treeview, a larger icon in a window, an editable word processing view, a print preview with different font metrics, a thumbnail presented when the mouse hovers over the icon). You add a controller to enable multiple interations in various views (selecting with the mouse, selecting by using tab key, activating an icon by double-clicking, activating an icon by pressing enter, modifying content by typing on the keyboard, copying and pasting content of the document, copying and pasting the document itself as an icon, dragging, dropping). Yes, in that complex world MVC can help to simplify the programming.
But the Web isn’t so complex. You have resources, which could be models in MVC. But you only have a handful of interactions and they’re all the same for every resource. You also have only a single way to view an item. You could possibly stretch the notion of representation to encompass multiple views but there’s nothing like the huge presentational differences that you get in a desktop GUI - just minor variations of markup language. MVC just obscures the mechanics of the Web, forcing the developer to write code that isn’t necessary and layer on more complexity when the mechanics start leaking through. Once you break your infatuation with MVC you find yourself writing much simpler code that mirrors the operations that drive the Web, and you actually start understanding a bit more about HTTP, the protocol that most frameworks try to pretend doesn’t exist.
Sep
22
2007
-
-
-
A loooong list I might add
-
Good reference for recovering Perl hackers. Exposes some of the more functional aspects of javascript.
-
An analysis of the algorithm that stumbleupon (may) use
-