another trueskill implementation; c++ with wrappers for python

This is a port of dougz/trueskill from python to c++;

You have two options:

1. run make in the root directory and launch the program via bin/runner
2. run python setup.py build_ext -i to create the .so file, and run the python test sample via python ptest.py

I timed each version via timeit while running the ps command on a 1s interval, and I included the last entry for each program. I used my ptest.py file, and a modified version of sample.py in the dougz implementation.

dougz pure python implementation:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
jbueski+ 15974  100  2.3 927368 433584 pts/8   Rl+  16:36   1:31 python sample.py

10,000 iterations in 91.2056720257s
 Alice: mu=33.208  sigma=6.348
   Bob: mu=27.401  sigma=5.787
 Chris: mu=22.599  sigma=5.787
Darren: mu=16.793  sigma=6.348

~110/s

this version:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
jbueski+ 19989  100  0.7 159044 131628 pts/17  R+   16:41   0:39 python ptest.py

1,000,000 iterations in 39.7102029324s
[(33.20778932686388, 6.347937213962031, 1),
 (27.4014978831654, 5.787057811781698, 2),
 (22.598576351284233, 5.787115941230072, 3),
 (16.793374093106948, 6.348053082283166, 4)]

~25,182/s

Above we can see a roughly 200x speedup from the c++ version.

I have not written any c++ in over 5 years. This port of a python program was an attempt to refresh my memory on writing c++. It is in no way optimized, and for that matter I can guarantee it is not the best way to implement it. I did however run it through valgrind and removed all memory leaks, so that's something.