Populations, samples, and machine learning
An experiment’s population is a complete collection of objects that we would like to study. These objects can be people, machines, processes, or anything else we would like to understand experimentally.
Of course, we typically can’t measure all of the population. Instead, we take a subset of the population—called a sample—and infer the characteristics of the entire population from this sample.
However, this inference that the sample is somehow representative of the population assumes the sample size is sufficiently large and that the sampling is random. This means selection of the sample should be such that no one group within a population are systematically over- or under-represented in the sample.
Machine learning is a field that makes extensive use of measurements and statistical inference. In it, an algorithm is trained by exposure to sample data, which is called a training set. The variables measured are called features. Typically, a predictive model is developed that can be used to extrapolate from the data to a new situation. The methods of statistical analysis we introduce in this chapter are the foundation of most machine learning methods.
Consider a robot, Pierre, with a particular gravitas and sense of style. He seeks the nicest pair of combat boots for wearing in the autumn rains. Pierre is to purchase the boots online via image recognition, and decides to gather data by visiting a hipster hangout one evening to train his style. For a negative contrast, he also watches footage of a white nationalist rally, focusing special attention on the boots of wearers of khakis and polos. Comment on Pierre’s methods.
Pierre must identify features in the boots, such as color, heel-height, and stitching. Choosing two places to sample certainly enhances the sample or training set. Positive correlations can be sought with the first group in the sample and negative with the second. The choosing of “desirable” and “undesirable” sample groups is an example of supervised learning, which is to say the desirability of one group’s boots and the undesirability of the other’s is assumed to be known.
Online Resources for Section 4.1
No online resources.