From dd470f2dbad7b4cf003014f188c688f6e8a3c2fd Mon Sep 17 00:00:00 2001 From: Leonid Boytsov Date: Thu, 5 May 2016 22:43:39 -0400 Subject: [PATCH] Update README.md --- previous_releases_scripts/README.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/previous_releases_scripts/README.md b/previous_releases_scripts/README.md index 38a2699..1e22553 100644 --- a/previous_releases_scripts/README.md +++ b/previous_releases_scripts/README.md @@ -3,6 +3,8 @@ Notes on reproducibility: * All scripts except for **vldb2015** should use binaries [version 1.0](https://github.com/searchivarius/NonMetricSpaceLib/releases/tag/v1.0). * **vldb2015** should use use binaries [version 1.1](https://github.com/searchivarius/NonMetricSpaceLib/releases/tag/v1.1). -Note that release 1.1 contains implementations of proximity graphs that are actually more efficient than those used for vldb2015 paper. In the case of light-weight distance, e.g., L2, the new implementations are about 3 times as efficient as the old ones. However, in the case of expensive distances the difference is much less. In particular, in the case of the normalized Levenshtein distance, the improvement is only 30%. Nevertheless, the old versions are still present under the names ``small_world_rand_old`` and ``nndes_old``. +Note that release 1.1 contains implementations of proximity graphs that are actually more efficient than those used for vldb2015 paper. In the case of light-weight distance, e.g., L2, the new implementations are about 3 times as efficient as the old ones. -However, the provided vldb2015 scripts will use **newer versions by default**, because it does not affect the big picture and conclusions of the paper. +However, in the case of expensive distances the difference is much less. In particular, in the case of the normalized Levenshtein distance, the improvement is only 30%. Nevertheless, the old versions are still present under the names ``small_world_rand_old`` and ``nndes_old``. + +The provided vldb2015 scripts will use **newer versions by default**, because it does not affect the big picture and conclusions of the paper.