A New Metric

Back in the 1980s, Roger Hockney invented two measures for computer performance. The first, r_∞, is the maximum possible performance when given infinite data (i.e., when startup overheads are ignored). The second, n_½, is how big the problem has to be for the machine to achieve half of its peak performance. Hockney observed that n_½ is effectively infinity in many cases, since no machine ever delivers more than 10-20% of the performance is manufacturer claims it will.

In the mid-90s, I suggested that another interesting metric would be how long it takes a programmer to learn how to write code that achieves half a machine’s rated performance. I called it p_½, and later revised its definition to be, “How long does it take a newcomer to learn enough about X to solve half of the problems they will ever use X to solve?” It turned out to be a poor metric, but not for the reason I originally thought. “Different people have different needs” wasn’t the problem: different people have different heights and blood pressures too, but that doesn’t stop us talking about means and distributions. The real problem is that the metric is unstable: the more people learn about a tool, the more problems they can solve with it, so the horizon recedes as they approach it.

So here’s my latest attempt:

w_½ is the time required for half of learners to believe that something was actually worth learning.

Based on purely personal experience teaching graduate students with no formal computer science background who want to do some data science:

I am currently trying to figure out what the likely w_½ values are for SQL, security, and ethics. If you have thoughts on these topics or others, please drop me a line.