Open BenLangmead opened 9 years ago
It turned out that the variability I was seeing was probably because of the empirical read/fragment length distribution and a problem with how it's used in the bias correction code:
https://github.com/cole-trapnell-lab/cufflinks/pull/32
But the random number generator seeding issue stands.
would there be any value in using the built-in RNG available with c++11 or newer standards? I think it provides for seed generation.
There seem to be many places where boost random number generators are instantiated (e.g.
boost::mt19937 rng;
insimulate_count_covariance
) but without a specified seed. So the default seed for the generator is used. Seems like they should be seeded with some function of the user-defined--seed
value. I noticed this because I see some isoforms for which the coverage estimate varies a lot when the BAM is perturbed slightly, but which do not vary when--seed
is varied.Ben