CNVM stochastic approximation is slow #25
Description
Currently sample_stochastic_approximation uses an inefficient version of Gillespie which requires the calculation of a O(M²) sum in each iteration and the sampling is also not efficient. (O(1) should be possible, but currently is O(log(M)).)
As a result, it is actually faster to use the full CNVM model with complete network, which is equivalent.