None of the transition matrix sampler tests do test whether the right distribution is sampled. At least the 1D and 2D density validations in the paper should be mapped to a test.
A general way of testing a Bayesian model is to fix a transition matrix P, then generate N trajectories of a fixed length from it. Using transition matrix sampling on each of the count matrices, we should find that on average, c% of the simulations lie in the c% confidence interval.
None of the transition matrix sampler tests do test whether the right distribution is sampled. At least the 1D and 2D density validations in the paper should be mapped to a test.
A general way of testing a Bayesian model is to fix a transition matrix P, then generate N trajectories of a fixed length from it. Using transition matrix sampling on each of the count matrices, we should find that on average, c% of the simulations lie in the c% confidence interval.