A reduction operation combines intermediate information from a homogeneous group of entries into a result. In MapReduce, mappers extract information and reducers receive entries grouped by category or key. section 13.2.2
A reducer can perform different work on each group:
- Count coin occurrences, sum their values, or filter entries. section 13.2.2
- In k-means, compute each cluster’s center of mass to update its centroid. section 13.3
$100: 1 50¢: 1 $100: 1 $1: 1 25¢: 1 ...
See sample mapper output listing coin occurrences before reduction. section 13.2.2
Grouping lets reducers work on separate categories. In canopy clustering, filtering centroids during reduction can leave some points outside every canopy, so the reducer’s role may affect coverage. section 13.2.2 section 13.3.3
