I would like to build a hierarchical model of sorts… essentially - I would like to sample

```
level_0 = Normal('level_0', mu=0, sigma=1, size=2)
mapping = [0,1,0,1,1]
mu = level_0[mapping] #this line clearly does not work - see below
level_1 = MvNormal('level_1', mu=mu, cov=np.eye(5)
```

what I’d like to do is sample the multivariate normal where the mean vector is:

```
[level_0[0], level_0[1], level_0[0], level_0[1], level_0[1]]
```

how can I efficiently construct this?