Drawing a Cholesky-decomposed correlation matrix

[Antonis_Mav]

The following works only when using LKJ_sd_dist=pm.Exponential.dist(1):

@chartl

sd_dist = pm.HalfNormal.dist(sd=1)
packed_chol = pm.LKJCholeskyCov(‘chol_cov’, eta=2, n=n, sd_dist=sd_dist)
chol = pm.expand_packed_triangular(n, packed_chol, lower=True)
cov = pm.Deterministic(‘cov’, tt.dot(chol, chol.T))
sd = tt.sqrt(tt.diag(cov))
corr = cov / sd[:, None] / sd[None, :]

The data is normalized to unit standard deviation. Do you think that the beta spike method can be applied to avoid using the script above?