Thanks to everyone who replied, (and special thanks to Alex Martelli,) I
was able to accomplish what I originally asked for: a shuffle() which is
both fast and without bias. It is the first time I try to optimize
python code, and it is definitely a very rewarding experience.
To bring closure to this thread, I will provide below some statistics
obtained from cProfile:
1. using random.SystemRandom
ncalls tottime percall cumtime percall filename:lineno(function)
4182879 35.154 0.000 211.006 0.000 os.py:724(urandom)
4182879 31.981 0.000 248.291 0.000 random.py:767(random)
40578 17.977 0.000 266.300 0.007 random.py:250(shuffle)
4182879 29.487 0.000 29.487 0.000 {posix.close}
4182879 99.460 0.000 99.460 0.000 {posix.open}
4182879 41.794 0.000 41.794 0.000 {posix.read}
2. my original optimization
8268 0.133 0.000 2.577 0.000 os.py:724(urandom)
4134322 15.094 0.000 21.606 0.000 baccarat.py:70(get2bytes)
4131441 17.221 0.000 41.722 0.000 baccarat.py:85(randrange)
40590 7.059 0.000 48.795 0.001 baccarat.py:112(shuffle)
3. using Alex's generator with string
4117 0.058 0.000 2.048 0.000 os.py:724(urandom)
4214795 10.186 0.000 14.880 0.000 baccarat.py:93(rand2)
4211790 8.883 0.000 23.763 0.000 baccarat.py:106(randrange)
40739 6.101 0.000 29.878 0.001 baccarat.py:131(shuffle)
4. using Alex's generator with array
2081 0.029 0.000 1.736 0.001 os.py:724(urandom)
4161569 1.724 0.000 3.473 0.000 baccarat.py:100(rand2new)
4158474 8.315 0.000 11.788 0.000 baccarat.py:113(randrange)
40514 5.726 0.000 17.528 0.000 baccarat.py:138(shuffle)
--
http://mail.python.org/mailman/listinfo/python-list
