Statismo's use of std::time() as a random seed is badly suited for concurrent processing

[normanius]

To seed random number generators, statismo uses the result of std::time(0). This is problematic because std::time(0) returns the time passed since epoch, measured in seconds. If multiple threads/process are started in very short time, the seed to the random number generator will always be the same, which is a problem if one wants to draw multiple samples from the same model/input in a parallel manner.

Of course there are ways to circumvent this issue, like ensuring that no two threads/processes are created in the same second. However, it would be more user friendly if the random number generator is initialized with more randomness.

Apparently, this is a bit more complicated than anticipated, but C++11 standard simplifies this considerably, as discussed in this article:

http://www.pcg-random.org/posts/cpp-seeding-surprises.html

[normanius]

If @muefra's pull request #268 is merged, we could simply replace std::time(0) with std::random_device{}(). This should be completely fine in the context of statismo, because the the sampling wth gaussian processes is so heavily input-dependent, that certain biases in the random number generator (see the link above) will play a negligible role.

[normanius]

There is a boost correspondent to std::random_device, but I'd prefer to use the latter directly.

https://www.boost.org/doc/libs/1_46_0/doc/html/boost/random_device.html