Conclusion

This chapter made uncertainty something we can compute with.

You learned that:

• a sample space contains possible outcomes
• an event is a set of outcomes
• probability rules keep uncertainty consistent
• random variables turn outcomes into numerical values
• distributions assign probabilities to values
• expectation is a probability-weighted average
• variance measures spread around the mean
• covariance measures co-movement, not causation
• independence means one event does not change another event's probability
• conditional probability changes the context
• Bayes' rule updates probability after evidence
• named distributions encode common modeling assumptions

The Main Skill

The main skill is to ask what uncertainty a probability statement is describing.

Is it about an event, a random variable, a label distribution, a loss value, or a belief after evidence?

That question prevents many mistakes. It keeps you from treating a model score as certainty, an expectation as a guaranteed outcome, or a covariance as proof of causation.

It also keeps probability tied to modeling assumptions. A distribution depends on the possible outcomes, the context, and the way probability mass is assigned. Changing any of those can change the answer.

What Comes Next

Numerical computation comes next.

Probability gives exact formulas. Numerical computation asks how those formulas behave when computers use finite-precision numbers.

Keep This Question Nearby

When a model outputs a probability, ask:

What uncertainty is this number describing, and what assumptions produced it?