models are simplified versions of reality.

DIMENSIONALITY REDUCTION Dimensionality reduction is a process you're already familiar with. Photography is an example; it reduces the three-dimensional world down to a flat, two-dimensional photo you can carry in your pocket. With datasets, we're working with rows and columns: observations and features. The number of columns (features) in a datase

unsupervised learning: a collection of tools designed to discover hidden patterns and groups in datasets when no predefined groups are available. It's a powerful technique used in a variety of fields, from segmenting customers into different marketing groups, to organizing music on Spotify or Pandora, and organizing the photos on your phone.

unsupervised learning. You didn't come in with preconceived notions about the data, but instead let the data organize itself.2

Doing data science: Straight talk from the frontline.

Whenever N is large, it's easy to make the leap and think N = ALL; every possible data point is at your disposal. But believing N=ALL does not absolve you from thinking about data quality and bias. (Remember the lessons in Chapter 4, “Argue with the Data.”) Are you truly capturing people from the population you care about?

In short, statistical inference follows these steps: Ask a meaningful question. Formulate a hypothesis test, setting the status quo as the null hypothesis, and what you hope to be true as the alternative hypothesis. Establish a significance level. (5% or 0.05 is an arbitrary but often-used number.) Calculate a p-value based on a statistical test. C

What is a false positive error? It's when evidence appears to confirm the reality of the alternate hypothesis when instead it should have been rejected (e.g., a man has a positive pregnancy test).

On the other hand, a false negative error happens when you accept a false null (e.g., a pregnant woman has a negative pregnancy test).