Variables and Observations

One day on a stroll downtown I came across a Ferrari on a surface parking lot and was immediately struck with the irresistible idea of a little social experiment. I took out my cell phone and my keys and walked up to the car as if it was mine and I was finishing up a conversation before getting in. People noticed quickly, they stared, took pictures. A crowd formed. Despite being on the phone I was interrupted twice, once by a man asking if I needed directions, and then by a pair of teenage girls who wanted to know if I knew someone whose name I couldn’t recognize so I winked and this drove them mad. I was a star, bathed in glory. Interestingly, no one asked me for spare change.

In numerical methods it is often not possible to ignore the limitations of the tools, especially when the tool is something as basic as floating point representation. A great deal of care needs to be taken to ensure that the implementation of a numerical algorithm will be accurate despite the limits of machine precision. If instead you try to think purely in terms of the algorithm, your results may not be what you expect. Witness my recent experience adapting a classic algorithm for calculating the Normal Quantile function into Python.

Sociology teaches the skills needed to analyze data and is thus of increasing signficance in today’s market.

There are several annoying phrases that come up from time to time in my work. I want to share some of the most egregious of business language failures. Note that this is only the beginning.

Last month I began hosting this website on a VPS provided by Linode.com and am happy to report being yet another happy customer of their excellent service. The entry level Linode 512 plan provides more than enough resources for a small to medium sized hobby or personal site. They have an excellent administrative system which makes provisioning a new VPS a simple painless operation. There is also an extensive collection of documentation at the Linode Library which includes guides for many of the common tasks you will encounter when setting up a new site. Their support is also outstanding. It is difficult to consider alternatives when Linode has already done everything right.

When I first read Paul Graham’s essay How to Make Wealth, I was quickly seduced, like, I am sure, many young corporate cubicle-dwellers who think themselves more clever than their employers, and found myself day-dreaming about quitting my job and forming a startup with a catchy url and the $50 Million I would get for selling it to Google. At the time, I was able to overlook the flaws and inconsistencies in the argument because the inspirational nature of the piece was so compelling. On closer examination, however, I find that underlying Graham’s motivational pep talks for the under-appreciated geek is a tenuous philosophical position that demands proper refutation. Whether this essay will succeed in righting Graham’s injustices remains to be seen. There are so many problems with Graham’s thinking that it is difficult to organize a focused response. Thus, with this essay, I intend mainly to introduce the reader to an alternate line of thought, lest they be so overwhelmed by Graham’s crusades that their mental capacities for logical reasoning are badly warped.

After 7 years as a customer of Pair Networks, I am leaving the safe comfortable confines of shared hosting for the wild jungle world of VPS. It was a difficult decision because Pair.com has an excellent setup and provides the best customer service and support in an industry where such features are hard to find.

A look at the kernel internals for handling file I/O operations.

In this series of posts, I will describe the technical foundations of a data analysis system that can support the goal of high-speed analysis of large datasets in real time. The focus will be on developing a system that can power a web application, using generic datasets. This means we cannot make any assumptions about the datasets that will be used, since we want a general purpose application that can work with a variety of data. This kind of optimization is not hard, but it can be tricky, and it will involve cost-benefit trade-offs for which real-world benchmarks are the best kind of success metric. In the end, we will have a basic blueprint for a data analysis system that can analyze data with blazing speed.

Today is a Palindrome Day for those who use the standard ISO date format of “yyyymmdd”. Enjoy it because there will not be another one for more than 8 years.