Serving the Quantitative Finance Community

 
User avatar
quantmeh
Posts: 0
Joined: April 6th, 2007, 1:39 pm

Let's benchmark C++, Java, C#

June 8th, 2010, 4:54 pm

QuoteOriginally posted by: renormI wasn't 100% right by saying that each thread can't have its own copy of mersenne twister. It turns out, that each thread can construct its own copy of RNG using a seed unique to that thread. Seeds must be properly "mixed", so that instances of RNGs don't generate overlapping sequences.this mixing is not against overlap, it's simply to have different seeds.
 
User avatar
renorm
Topic Author
Posts: 1
Joined: February 11th, 2010, 10:20 pm

Let's benchmark C++, Java, C#

June 8th, 2010, 6:14 pm

Not exactly. Seeding get's you different sequences. Seeding with mixing get's you non-overlapping sequences. Mixing ensures that the states resulted from two different seeds are separated by extremely large number of transitions.
 
User avatar
gsgiles
Posts: 0
Joined: December 13th, 2008, 1:39 pm

Let's benchmark C++, Java, C#

June 8th, 2010, 6:20 pm

I would get my random seeds externally see: random.org or hotbits.orgYou can use something like truecrypt to build a large file of random bits using these seeds, alternatively you can buy lots of bits from random. Machine generated are never as random as these.My $0.02
 
User avatar
renorm
Topic Author
Posts: 1
Joined: February 11th, 2010, 10:20 pm

Let's benchmark C++, Java, C#

June 8th, 2010, 6:35 pm

MC needs deterministic sequences which pass various randomness tests.