useR! 2024

Optimization benchmark sets are commonly used to evaluate the quality and speed of optimization solvers. These problems are typically collected from real world applications. We suggest using statistical models to automatically generate optimization problems. This has the advantages that for statistical models the data generating process is typically well known therefore it is easy to generate data for the model and then transform the data into an optimization problem. Furthermore, for statistical models, properties like convexity and unboundedness are typically well known.

Group-wise aggregation is one of the most common operations in data analyses.. There are use cases where the grouping is determined dynamically by collapsing smaller subsets into larger ones, to ensure sufficient support for the target aggregate. Examples include cases where some of the target groups suffer from missing data, or cases where the quality of target group data is judged to be too low. Often, hierarchical classifications serve as a basis for forming larger groups, but custom 'collapsing schemes' are in use as well. In this presentation we demonstrate the R package 'accumulate' [1] that offers interfaces for defining grouped aggregation, where the grouping may be dynamically determined, based on user-defined aggregations, user-defined decision rules, and user-defined collapsing schemes. The package offers several ways to define collapsing schemes, including tabular definitions that can be maintained separately from the aggregation code. It also includes facilities to use hierarchical classifications and for testing the (possibly complex) decision rules that user can create. [1] https://cran.r-project.org/package=accumulate