Version 1.9.2 * Removed redundant assignment (issue #422 @fluber) * Removed unnecessary null checks before delete (issue #420 @elfring) * Reverted PR #424 due to lack of portability * fscanf fix (PR #467 @rdelfin) * Out of bounds check (modified PR #455 @legrosbuffle, also included in PR #319 @seth-planet) * Fixed MacOS build (PR #470 @johnhe4, modified for linux) * Fixed build system with dummy.c hack (PR #457 @pemmanuelviel) * Fixed misleading indentation in util/any.h (PR #365 @psteinb, also PR #428, #430, #459) * Included datasets in repo * Correct typo in definition (PR #419 @SergioRAgostinho) * Fix typos (PR #279 @gadomski) * CMakefile CUDA sources fix (PR #458 @pemmanuelviel) * Documentation fix (PR #456 @pemmanuelviel) * Scoping issue fix (PR #405 @greenbrettmichael, also PR #469, issue #386) * Documentation fixes (PR #460 @pemmanuelviel) * Changed return value (PR #461 @pemmanuelviel) * Fixed CUDA crash - guarantee prealloc > 0 (PR #437 @neka-rat) * Fixed wrong variable use (PR #433 @XinyiYS) * Fixed RNG initialization (PR #424 @SiddhantRanade) * Updated link to PDF (PR #474 @kleinma) Version 1.6.11 * bug fixes Version 1.6.10 * fixed a radiusSearch bug introduced in 1.6.9 Version 1.6.9 * bug fixes * fixed radius search bug on MSVC compiler * fixed windows linking problems Version 1.6.8 * bug fixes, low dimensional search speedup Version 1.6.7 * bug fixes Version 1.6.6 * misc bug fixes Version 1.6.5 * fix compilation problem on some C++ compilers * fixes in the python bindings Version 1.6.4 * small bug fix Version 1.6.3 * radius search speedup Version 1.6.2 * slight API changes to the C++ bindings, now the main index type is templated * on the distance functor which makes it easier to use custom distances * new kd-tree implementation optimized for low dimensionality data * experimental MPI support for cluster computing Version 1.5 * new C++ templated API * saving/loading of indices to disk * threadsafe search * new distance types (thanks to Radu Bogdan Rusu and Romain Thibaux for the patch) * (api change) autotuned is no longer selected by passing a precision >0, it's used when the algorithm type is set to autotuned