prediction & decision
Chapter 4
Once we have computed probabilities, we want to use them:
Predictive modeling allows us to expand static theories into the temporal domain, predicting changes in behavior as internal (bodily) or external environmental conditions change over time. These equations underlie the powerful data science of realtime predictions provided by 'big data'. Decision theory allows us to optimize choices, even when there is uncertainty about outcomes. This involves combining the information conveyed by probabilities with values, which then determines the decisions we make within a given probability landscape. As an important example, we demonstrate that the values we attach to errors of different types determines whether we report the mean, median, or peak of a probability distribution as its location measure.
Programming Asides:

binomial prior predictive and sampling [p189]

negative binomial prior predictive and sampling [p190]

gaussian prior predictive and sampling [p193]

gaussian prior predictive with unknown dispersion and sampling [p194]

binomial with sampling [p198]

normal with known variance and sampling [p201]

cell division [p204]

limit of zero weight given to previous temporal intervals [p215]

memory distortion based on previous temporal interval [p218]

zeroone loss/risk [p236]

linear loss/risk [p240]

squarederror loss/risk [p243]

the 'unbiasedestimator' fallacy [p247]