#ordered-data
Learned Index Search
Learned Index Search Learned index search treats an ordered index as a prediction problem. Given a key, a model estimates the position where that key should appear in sorted data. The algorithm then corrects the estimate by searching a small range around the predicted position. The main idea is that a sorted array defines a cumulative distribution function from key to position. If this function is predictable, a model can...
Recursive Model Index Search
Recursive Model Index Search Recursive model index search uses several learned models arranged as a hierarchy. A top model selects a lower model, and the lower model predicts the likely position of a key in sorted data. The final prediction is then corrected by searching inside a bounded local range. This is the main search procedure behind a recursive model index, often shortened to RMI. It replaces part of a...