Hi, I’m not quite sure what your end goal is. Do you wish to model something that is univariate or multi-variate? If Univariate then you can use pm.Normal() rather than pm.MvNormal. if multi-variate then you can use pm.MvNormal() but set D>=2.
chol parameter in MvNormal should be the cholskey matrix which when multiplied by itself transposed gives the covariance matrix. You are inputting the covariance matrix so you should use cov parameter. (though in your code this doesn’t matter since you just have non-zero diagonals)
Also you are inputting an observed without any variance, I’m not sure how well PyMC3 will handle that. If you are just testing things out I suggest you use pseudo random numbers from np.random.