Using MCMC based model in closed loop

Questions
14

Looking at this, the critical flaw seems to be that it assumes that the variables to be updated – alpha, beta0, and beta1 – are independent.

I’m not sure we’re looking at the same thing. This block:

    blocks, mars, zscores = transform_variables(trace, varnames)
    # compute the correlation of transformed variables
    pcor = np.corrcoef(zscores.T)
    print(pcor)
    L_p = np.linalg.cholesky(pcor)
    pr_lat_z = pm.Normal('pr_copula_z__', 0., 1., shape=(mars.shape[1],))
    pr_lat = pm.Deterministic('pr_copula__', tt.dot(L_p, pr_lat_z))

explicitly models the correlation of parameters from the traces.