Spatial approximation + sequential scan = efficient metric indexing

Proximity queries in metric spaces, or metric proximity queries for short, have instances suffering from the so-called "curse of dimensionality" referred to the apparent impossibility of beating the brute force approach (an exhaustive search to satisfy a metric proximity query). In other words, the cursed spaces resists indexing. Current research tends to alleviate this particular situation, since easy spaces can be indexed neatly. A recent hypothesis establishes that it is just a fraction of the database which is resilient to indexing, and the proper approach would be to index each part separately. This is precisely the approach of this work. We use two different techniques for indexing complementary parts of a metric database. The dynamic spatial approximation tree (dsa-tree) is a recently proposed data structure for searching in metric spaces. It has been shown that it compares favorably against alternative data structures in spaces of high dimension or queries with low selectivity On the other hand, pivot-based algorithms are effective tools for proximity searching in any metric spaces, if sufficient storage space is available. In this work we combine the concepts of spatial approximation and pivot-based algorithms. The new data structure maintains the full features of a dsa-tree while using the available memory to improve the query time. We show experimentally that our data structure is competitive choice for searching in cursed metric spaces.