New Book from O'Reilly: Programming Collective Intelligence(radar.oreilly.com)
radar.oreilly.com
New Book from O'Reilly: Programming Collective Intelligence
http://radar.oreilly.com/archives/2007/08/programming_col.html
18 comments
I could do without the Web 2.0 hype, but the table of contents does look very interesting: it's basically an introduction to a particular subset of machine learning/data mining. I'd love to see more focus on solving these kinds of hard problems, and less on rounded corners and AJAX, among the "Web 2.0" crowd.
Exactly! "Web 2.0" has somewhat of a bad image because people think of the flash instead of the substance. "Web 2.0" actually has a lot of deep technology to it; I think "harnessing human intelligence" is a technical trend that will continue for decades. The hard part is figuring out unique, useful ways of algorithmically working with user generated data, which in turn would lead to new categories of applications.
I still think Peter Norvig's "Artificial Intelligence" book plus a good book on Data Mining (someone suggested Frank and Witten above) might deliver more value than this.
As Henning pointed out above , there aren't enough pages to cover even a few of the topics properly. It looks like the coverage is fairly thin and lacking in depth, and is more of a "some basic AI algorithms for the hot topic of the day" type book.
I haven't seen the book yet but the table of contents makes me suspect the quality. "Support Vector Machines" for example are fairly complex beasts. The book has one section (from what I can see on the TOC) about SVMs.
Norvig's AIMA is a brilliant book precisely because it lays a solid foundation for so much of AI. And it is still a very hefty book. I somehow doubt that this is of the same caliber.
But hey, I could be wrong.
Just a personal opinion and not to dampen anyone's enthusiasm.
As Henning pointed out above , there aren't enough pages to cover even a few of the topics properly. It looks like the coverage is fairly thin and lacking in depth, and is more of a "some basic AI algorithms for the hot topic of the day" type book.
I haven't seen the book yet but the table of contents makes me suspect the quality. "Support Vector Machines" for example are fairly complex beasts. The book has one section (from what I can see on the TOC) about SVMs.
Norvig's AIMA is a brilliant book precisely because it lays a solid foundation for so much of AI. And it is still a very hefty book. I somehow doubt that this is of the same caliber.
But hey, I could be wrong.
Just a personal opinion and not to dampen anyone's enthusiasm.
I agree completely. Web 2.0 isn't about flashy web graphics it's about understanding how to bring users together in such a way that you create something new and exciting. Data-mining is the best way to generate actual businesses from 'Web 2.0' anything that helps people get their feet wet in this complex field is worth while.
I agree, AJAX really is only a tool for whats actually going on in the web right now. A sophisticated website is not one where ajax is at the core, I think its one where everytime the user visits and actively participates with the site, the user experience gets better the next time they come.
"the defining moment of the Web 2.0 revolution was Google's invention of PageRank"
Huh? PageRank dates from the late 90s. And it's implemented in a compiled language (C++, I think -- not even GC'd).
And it involves elementary linear algebra. Heresy!
Now, as for the book itself. It looks like it tries to cover waaay too much -- support vector machines and other forms of supervised learning, unsupervised learning/clustering, optimization and evolutionary computation, applications to collaborative filtering, along with using particular libraries. There's no way you could cover all that in 358 pages!
I gave a talk on evolutionary computation to a Ruby users group and their jaws literally dropped. Web people don't give a shit about something if it doesn't involve HTTP, a programming language that's currently en vogue, JavaScript, or relational databases (really only MySQL or maybe Postgres).
If you're interested in an introductory data mining book with a practical focus, may I instead suggest Witten and Frank's "Data Mining: Practical Machine Learning Tools and Techniques"? It uses Weka throughout, which is mature and nice.
Huh? PageRank dates from the late 90s. And it's implemented in a compiled language (C++, I think -- not even GC'd).
And it involves elementary linear algebra. Heresy!
Now, as for the book itself. It looks like it tries to cover waaay too much -- support vector machines and other forms of supervised learning, unsupervised learning/clustering, optimization and evolutionary computation, applications to collaborative filtering, along with using particular libraries. There's no way you could cover all that in 358 pages!
I gave a talk on evolutionary computation to a Ruby users group and their jaws literally dropped. Web people don't give a shit about something if it doesn't involve HTTP, a programming language that's currently en vogue, JavaScript, or relational databases (really only MySQL or maybe Postgres).
If you're interested in an introductory data mining book with a practical focus, may I instead suggest Witten and Frank's "Data Mining: Practical Machine Learning Tools and Techniques"? It uses Weka throughout, which is mature and nice.
Just because people are interested in "Web2.0" doesn't mean they're not going to be interested in the kind of stuff in this book.
In my opinion, interest in this sort of material is going to be propelled forward by Web2.0. It's not at all true that web programmers "don't give a shit" about mathematical modelling, etc. It's just that there hasn't been much accessible code for people to look at (much of it tangled in obscure, obtuse, proprietary academic and military labs).
It's true that this book seems to be attempting to cover a lot of topics, but that's probably the point: a decade ago, there was no popular concept, let alone interest, in relational databases. That changed because the street found a use for it, as the saying goes. Anyway, I'll second your Witten and Frank recommendation, cool book.
In my opinion, interest in this sort of material is going to be propelled forward by Web2.0. It's not at all true that web programmers "don't give a shit" about mathematical modelling, etc. It's just that there hasn't been much accessible code for people to look at (much of it tangled in obscure, obtuse, proprietary academic and military labs).
It's true that this book seems to be attempting to cover a lot of topics, but that's probably the point: a decade ago, there was no popular concept, let alone interest, in relational databases. That changed because the street found a use for it, as the saying goes. Anyway, I'll second your Witten and Frank recommendation, cool book.
It looks like a stripped down version of AIMA (http://aima.cs.berkeley.edu/) with two differences: It doesn't cram everything into an "agent based approach" and it gives usable examples right away skipping theory (no 100 pages devoted to first order logic).
In the sample text they mention Hot or Not, Google, Amazon, Netflix, and a few other companies to quickly give a "real world" view of what is useful. The book certainly will get more people interested in doing more sophisticated computations on their data. (Not that I think AIMA using Romanian cities or endless "sue is pat's mother. sarah is pat's daughter => sue is sarah's grandmother" examples make the material seem less useful.)
In the sample text they mention Hot or Not, Google, Amazon, Netflix, and a few other companies to quickly give a "real world" view of what is useful. The book certainly will get more people interested in doing more sophisticated computations on their data. (Not that I think AIMA using Romanian cities or endless "sue is pat's mother. sarah is pat's daughter => sue is sarah's grandmother" examples make the material seem less useful.)
It doesn't seem like a "stripped down version of AIMA" at all, IMHO. AIMA is an AI book, this is an introductory data mining book. As such, this book talks about algorithms for clustering, classification, feature identification, collaborative filtering, etc., none of which are really addressed in depth in AIMA. There is a relatively brief section on learning techniques in AIMA, but it doesn't go into much depth, and is more focused on reinforcement learning than on typical data mining techniques (automatic classification, clustering, etc.)
Does the book go into any depth on these issues or is the coverage superficial? Is there code? Thanks in advance
Its not out yet. I bought it on Amazon and then they told me a couple days later that it would take until November to actually receive it. I cancelled the order.
It's out-- see Tim O'Reilly's comment (currently the last one):
For those of you wondering whether to buy from Amazon or directly from O'Reilly, I heard from our Amazon sales rep that Amazon is temporarily out of stock, and is in fact showing as "not yet published." She wrote in email:
"Since I've heard from several of you regarding Programming Collective Intelligence and the status on Amazon, I thought I better send out a quick little note to explain the "glitch." As most of you have seen, Amazon's detail page for Programming Collective Intelligence is now showing as a pre-order but just last week it was "available." ... Here's what happened as it's been explained to me. Apparently, Programming Collective Intelligence ran out of stock as quickly as it was received in. Because it ran out of stock so close to the expected pub date, the system threw it back into a pre-order status."
For those of you wondering whether to buy from Amazon or directly from O'Reilly, I heard from our Amazon sales rep that Amazon is temporarily out of stock, and is in fact showing as "not yet published." She wrote in email:
"Since I've heard from several of you regarding Programming Collective Intelligence and the status on Amazon, I thought I better send out a quick little note to explain the "glitch." As most of you have seen, Amazon's detail page for Programming Collective Intelligence is now showing as a pre-order but just last week it was "available." ... Here's what happened as it's been explained to me. Apparently, Programming Collective Intelligence ran out of stock as quickly as it was received in. Because it ran out of stock so close to the expected pub date, the system threw it back into a pre-order status."
I tried to get it from Amazon again. Lets see if they can handle it this time ;-) I'm not going to buy it from Oreilly b/c I get free (amortized) shipping from Amazon and its $13 cheaper.
FYI, my order from Amazon shipped today (I ordered it 5 days ago).
And there are even used ones available on Amazon.
Holy cow - this could be really good. The content preview is giving me a feeling of wordiness though. Next time I'm in a bookstore I'll thumb through it to see if it's succinct enough.
I thought the same, however it still may be useful as a survey book. It's something I might read as a starting point. Sometimes it's discouraging to slog through a technical paper only to discover that it's not going to help you much.
try it on safari.informit.com free for 14 days.