I have a problem in which I want to infer the correlation structure of some data. I am using the LKJ distribution and I want to retrieve the posterior distribution of etas. However, the results always generate the same distribution. I have tried with many examples of generated data and I do not know what I am missing. For example, assuming that the correlation matrix is a Identity matrix, I am generating the most extreme case, where eta should tend to infinity. However, I find:

```
import pymc3 as pm
import numpy as np
D=10
y = np.identity(D)
il1 = np.triu_indices(D,1) #LKJCorr reads only the upper tri part of the matrix
obs = y[il1] # Taking upper tri of matrix
with pm.Model() as model:
h_eta = pm.Uniform('eta', lower=0.00001, upper=3000)
packed_L = pm.LKJCorr('obs_cov',eta=h_eta, n=D, observed=obs)
#step = pm.Metropolis()
trace = pm.sample(9000,njobs=2,nuts_kwargs=dict(target_accept=.95))
pm.traceplot(trace)
```