Introduction to Data Analysis

Online course material.

Peer-to-peer reputation system

Aura. Interesting.

Ideas for small data projects

Here are some ideas. Here are some more (medically related).

In data science, why Python?

A retrospective. The short answer is basically that MATLAB changed their licensing to exclude the Fraunhofer Institutes (possibly unintentionally), and Fraunhofer people responded by rolling their own.

Interesting point made: if it takes minutes to load up your dataset, it's nice to be able to work with the heap in a dynamic manner, adding new functions to work with the data structures already in memory. Python has ways of making that easy, and of course MATLAB had that rolled in from the start.  Any REPL language can do that.

This is essentially a question of in-memory indexing of a database.

Reservoir sampling

A cute little post about interviewing for work in data science, especially talking about reservoir sampling.

Wikidata

I'll just leave this here. Lots and lots of data.

Datasets released by Google

Google has actually released a lot of interesting ML datasets. Here's a short list.

Probabilistic programming languages

Apparently I was on a real language-design tear in April, too - here's a post on probabilistic programming languages, proposing semantic primitives for, well, probabilistic programming. Where do DSLs stop and plain old programming languages start? ... Good question.

I have to say, the BUGS language [that was to the old WinBUGS: here's OpenBUGS, the current project] looks pretty darned interesting - you're really using this to set up a model in a declarative manner, then invoking an engine that writes the "query results" into the original file, looks like.  I really like the cut of that jib.

Then there's Church. Wow. I think this might have Hofstadterian implications, honestly.

A practical intro to data science

Here's a good, link-rich article for ya.

Calculating rolling cohort retention - with SQL

This kind of trick is great stuff. I don't even know how to categorize it. Well, "data science", of course, but this general kind of algorithmic sleight-of-hand is always attractive.

Unix command line for data science

Here's a useful post.

Open street maps data

Openstreetmap.org provides the same service as Google Maps, but as an open-source project. Very, very nice! Listed under data science because I don't have a better place for public databases.

Python data tools

O'Reilly: "they keep getting better and better".  Yeah.  I'm gonna have to reimmerse myself in Python, I guess.  *snf*  I'll miss you, CPAN!

So you want to be a data scientist

Ha.  As long as we're on curricula.

Free data science books

A list of free online data science books, posted by a data scientist.  This is nice.

Blekko crawl data on Common Crawl

Web crawl data.  Also, Common Crawl is cool.

Blaze: compiling data science in Python

Blaze is the new - compiling - generation of NumPy.  It's fast.  I wonder why people don't just write macros that write C anyway?  (I'm thinking code quality, but seriously - that would be interesting to investigate.)

How to learn data science

Version 1.38 - nice list of resources.  Including linear algebra at OCW.

Smaller samples -> greater variation

DeMoivre's law - and how people really don't get it. This is really a good read.

Data exploration

Data exploration in Unix.

But a few days later, the O'Reilly "Try R" class at CodeSchool was posted on HNN and prompted a rich outpouring of data science linking (and book recommendations, too).  MIT OpenCourseWare statistics class, and an online book on statistics and R, stand out.  Also the recommendation not to learn R without a simultaneous good grounding in statistics itself, because R is first and foremost a statistics platform.