Building a Sample Dataset

A challenging dataset for testing a numerical prediction algorithm should have a few properties that make the dataset more difficult to make predictions from. If you are looking at TVs, it’s easy to infer that bigger is better, and such problems can more easily be solved with traditional statistical techniques. For this reason, it’s more interesting to look at a dataset where price doesn’t simply increase in proportion to size or the number of characteristics.

In this section, you’ll create a dataset of wine prices based on a simple artificial model. The prices are based on a combination of the rating and the age of the wine. The model assumes that wine has a peak age, which is older for good wines and almost immediate for bad wines. A high-rated wine will start at a high price and increase in value until its peak age, and a low-rated wine will start cheap and get cheaper.