dncrane / cover-tree Goto Github PK

Hi to all,
looking at the code frequency, I am wondering if this repo is dead? Did people move on? Is there another good implementation of a cover tree out there? Or should we fork/clone and continue letting this baby live??
Best,
P

I just replaced all the use of std::pair with using std::tuple and got a 30% speed improvement on insert calls:

--------------------------------------------------------------
Benchmark                       Time           CPU Iterations
--------------------------------------------------------------
BM_insert/100             4009190 ns    3993675 ns        175
BM_insert/200            14098351 ns   14066840 ns         50
BM_insert/500            79044530 ns   78869829 ns          9
BM_insert_improved/100    5105467 ns    5094695 ns        138
BM_insert_improved/200   18299317 ns   18259001 ns         38
BM_insert_improved/500  105163774 ns  104951121 ns          7

@DNCrane are you interested in such improvements and hence interested in moving to a more recent standard?

Have you considered adding an open-source license to the project so that the code can be used by others? If no license is given, the default is 'exclusive copyright by author' which means that nobody can legally use it....

I personally prefer the Apache APL 2.0 license for my own projects (for example, it provides patent grant and a patent retaliation clauses while being less restrictive than others, such as GPL), but others (MIT license, ...) are also good. See short license overview.

It would allow me to translate your code to Java and include it in my spatial index library (giving due credit to you, of course).

If you like, I can provide a pull request.

I am offering to port the tests to use the header-only catch(2) framework. I am suggesting to move to c++11 at least. Any objections/thoughts/feedback? If not, I'll send in a PR.

it would be useful to add a kNN method that returns not only points, but also their distances, e.g. as std::vec<std::pair<double, Point>>.