A critical analysis of recursive model indexes

The recursive model index (RMI) has recently been introduced as a machine-learned replacement for traditional indexes over sorted data, achieving remarkably fast lookups. Follow-up work focused on explaining RMI's performance and automatically configuring RMIs through enumeration. Unfortunately, involves setting several hyperparameters, the enumeration of which is often too time-consuming in practice. Therefore, this work, we conduct first inventor-independent broad analysis with goal understanding impact each hyperparameter performance. In particular, show that addition to types layer size, error bounds search algorithms must be considered achieve best possible Based our findings, develop simple-to-follow guideline RMIs. We evaluate by comparing resulting number state-of-the-art indexes, both learned traditional. simple sufficient competitive other whose configuration was determined using an expensive procedure. addition, while carefully reimplementing RMIs, are able improve build time 2.5x 6.3x.