andreaferretti / kmeans Goto Github PK

The java version of this k-means benchmark code is obviously cheating compared to other object-oriented language versions, i.e the ruby one. Since I'm not a language expert, I did not check other language versions other than java and ruby.

• The java version reused the Centroids array between iterations for every run loop, while the ruby version re-allocates.
• The java version uses addToThis() method to avoid allocating new Point object, while the ruby version allocates new Point object on every + operator calling. This makes the ruby version allocating Point object about 100,000 × 15 more times than the ruby version for every run. The divideToThis() does the same thing, but is fare less important then the former.

I can see the data available in points.json, is there any way I can increase the size of this file with adding more points ?

I find this experiment interesting. I think these are some interesting languages to try:

• Julia
• Idris

Andrea, this is a neat project. I wonder if adding a Dockerfile to each directory would make this project fun for you again, and relieve you of having to manually manage a polyglot host environment. If all tests are run in a container, it would be a fairly level playing field, and the results would still work as a relative comparison.

Hey, So I am using AWS cuda instance and trying to run cuda version. AWS Instance has everything installed and I have installed jansson from http://shellgui-docs.readthedocs.io/en/latest/gettingstarted.html successfully. Still I am getting this error. Any suggestion ? Thanks

./kmeans.out: error while loading shared libraries: libjansson.so.4: cannot open shared object file: No such file or directory

edit:

echo $LD_LIBRARY_PATH

/root/lib/

and /root/lib/ has following files:
libjansson.a libjansson.la libjansson.so libjansson.so.4 libjansson.so.4.9.0 pkgconfig

Hi, somehow I hit across your benchmark and got interested. Your C code scores so low which got me interested (doesn't nim compile into C so how can the native C be slower than nim?).

Any way, I wrote my own quick version of bench code and would like you to time on your machine.

https://github.com/hzhou/kmean.

You can directly compile the c file

gcc test.c && ./a.out

I developed the code in my own meta-layer (MyDef), so if you want a more readable version, you may read test.def directly.

I didn't use any json library nor hash maps, so no dependency and the parsing overhead should be negligible for the benchmark anyway.

Regardless the result, nice work!