> > Wow, this looks very relevant to Lucene! Could this index be used for > faster implementation of our skip lists? Even though they are static > (computed once at segment-write time) vs dynamic/online that these learned > indices are also able to handle, it looks like learned indices are still > better than simple binary search, and quite a bit more compact, for static > cases.
I do hope this index can be used for skip lists! Our "best linear fit" approximation to compress monotonic longs > (DirectMonoticWriter/Reader) looks like a simple example of these learned > indices too. Yes, this is the same linear approximation technique. With the concept of an optimal (minimal) sequence of segments and a hierarchy of segment layers, the paper gives a very compact yet precise approximation. >