Thank you a ton for your help, Jonah! It worked out well for me now, so helps settle what I want above:)
Unfortunately, I need to change the modeling part a little bit due to the nature of a project I am working on. When I did make a change on changing
sigma_y from a vector to a matrix, and re-ran my program, it gave me the error message:
normal(A*yT+epsilon, sigma_x) cannot take on the form [vector, matrix].
The reason this problem arises is because each of our parameters
x are **vectors **, and so the variance of
yT, which is
sigma_y, theoretically must be a co-variance matrix with size
[nrow, nrow]. Similar,
sigma_x should also be a co-variance matrix with size
Question: I wonder if you (or anyone in this forum) have encountered a similar model, where we are basically given the normal distributions of vectors, rather than random variables, and try to make Bayesian inference for the posterior distribution in the same way as my model in the original post? If anyone did, could you show me how you overcame the problem of having two vectors following normal distributions, which have different dimensions for their corresponding means and variance.