1. Defining costs#
Everything in Empulse starts from one object: a CostMatrix recording what
each of the four classification outcomes is worth. Metrics score with it, models train on it,
samplers resample by it and the bundled datasets ship one.
This stage covers writing that matrix down, handing it to an estimator, and getting per-row cost values safely through a pipeline. Once you have a cost matrix, Measuring in money turns it into a number.
The four outcomes, the cost/benefit sign convention, and the builder API — symbols, aliases, defaults and stochastic terms.
Plain fp_cost/fn_cost values or a full Metric as loss: which to use, and
the rules that govern both.
Metadata routing carries per-instance values through pipelines, cross-validation folds and hyperparameter searches.
Complete examples built from business parameters, including uncertain parameters and mixture distributions.