All writing

Predictions smedictions

How I improve my boosted tree using a plasma globe.

Predictions are hard, especially when it’s about the future. The first thing you want to ask when tasked with forecasting: is the thing you are trying to predict predictable?

Predictability boils down to the ability to spot patterns, based on things that we have seen before. If nothing connects the quantity in the future and the past, or there are no grounding rules (like laws of physics) dictating the process, then there’s no point forecasting those things.

Predicting something unpredictable (in principle) is just fortune telling, it hits and misses. (By the way, <bitchy> I’m a Libra everybody, so you know I hate being alone </bitchy>.  Btw btw, it’s prime-time false prediction, shame on you Cosmopolitan mag ). In that case, sophisticated models and metrics are just like cards and plasma globes. They’re there just to fool you. If there are people trying to sell you predictions like those, they are either being possessed by something (ghosts, or even scarier, money) or just plainly think that you are dumb. Or both.

So what constitutes things that are predictable ? Well, if there is an underlying principle ruling over it, then maybe. It could be physics or other laws of nature, distributional laws. There are processes whose mechanics are well defined, but it is still hard to predict the actual outcome, like motion of particles in liquid.

In the end, how good the prediction is depends on the scale and scope of the problem, and the variance of the variable at hand. And good is subjective, it’s relative to your expectations.