回复 第11楼 的 enthumelon:Are you sure you are comparing exactly the same algorithm? For example, if WinBUGS does not know how to sample directly, an adaptive rejection algorithm might be used. Are you implementing the ARS in C++ when doing speed comparison?
I dont' understand "并行是一个方面". What other benefits can we get from multi-threadding?
Actually, I heard that R is moving toward re-writing more code with OpenMP.
I wonder where is the ultimate cause of slowness in R. Is it parsing/interpreting overhead? Did you try to compile the R MCMC function before comparison?