O'Reilly Strata

Yes, looks cool. I'd suggest adding web and desktop versions of a little CRUD app - maybe a contact manager. Test suites would definitely be a good addition. There are interesting MOOC elements to this. it'd be very cool to collect all the implementations and analyze the different ways people solve the problem. Stanford had dinner some interesting work like this: http://strata.oreilly.com/2013/10/semi-automatic-method-for-grading-a-… Finally, it'd be great to open-source this. Set up a repo with assignments and test suites, then students fork the repo and code up the exercises. This would allow you to easily see how people are solving the problems, and make it easy for people to propose new projects and tests.

This is a cool tutorial! It's ironic that deep neural networks have become the biggest machine learning breakthrough of 2013: they were also the biggest machine learning breakthrough of 1957. The idea dates back to the Perceptron, one of the oldest ideas in AI. One thing to note: although there was a lot of initial excitement about Restricted Boltzman Machines, Auto-encoders, and other unsupervised approaches, the best results in the last year or so have all used conventional the back-propagation algorithm from 1974, with a few tweaks. http://en.wikipedia.org/wiki/Backpropagation Ben Lorica wrote a good article on the latest deep learning research from Google, and what's changed since neural networks were last popular in the 1980's: http://strata.oreilly.com/2013/10/deep-learning-oral-traditions.html What's old is new again.

For Panjiva: http://strata.oreilly.com/2012/12/panjiva-government-data-platform.htm…

There have been some posts on Hacker News describing this philosophy (see below, can't find the HN comment links unfortunately). Another advantage is that it encourages separating your policy from mechanism. For example, you can make your sampling rate or number of buckets a command line flag, and have all your parameters in the shell scripts, rather than having various constants strewn about the code. And obviously you can use each language where it's stronger. Python is better (and much faster) for the data cleaning part, whereas you might want to use ggplot in R. http://jeroenjanssens.com/2013/09/19/seven-command-line-tools-for-data… http://www.drbunsen.org/explorations-in-unix/ http://strata.oreilly.com/2011/04/data-hand-tools.html

I too am frustrated by the current selection of Quantified Self gadgets. My wishlist of features for a step counter: 1. Self-charging. The Misfit Shine gets close by using a watch battery that lasts months. 2. Self-syncing. 3. API, of course, so it's easy to automatically do things with the data (like send it to http://beeminder.com !). 4. Waterproof, so I don't have to remember to take it off in the shower. Basically I want something I can wear and never have to think about at all. More generally, I think what's missing (extreme bias warning) with Quantified Self is that it needs to evolve toward Programmable Self: http://strata.oreilly.com/2012/01/programmable-self-motivation-hacks-d… In other words, the point of collecting data about yourself should be to optimize yourself, especially fixing procrastination and other forms of akrasia like eating too much and moving too little. Sometimes just seeing your data is a nudge in that direction but in my opinion (did I mention my extreme bias as a cofounder of Beeminder) it mostly isn't.