Not sure if this is a bug.

However, I’ve been trying to sample a Negative Binomial model but constantly running into logp = -inf.

I have checked that there are no negative values anywhere, so not sure why this is happening.

Here is the code:

```
with pm.Model() as quantity_model_1:
alpha = pm.Uniform("alpha", lower=0, upper=10)
# Priors
sigma_a = pm.HalfCauchy('sigma_a', 1)
# Random intercepts
a = pm.HalfCauchy('a', beta=sigma_a, shape=no_stores)
b = pm.HalfCauchy('b', beta=1, shape=no_dow)
d = pm.Normal('d', mu=0.003, sd=0.001,testval = 0.05)
k = pm.Deterministic('k',var=b[dow]*pm.math.exp(d*temperature)) ##Note I specifically use exponential function to avoid any negative values.
y_hat = pm.Deterministic('y_hat',var=a[storeno]*k)
# Data likelihood
vol = pm.NegativeBinomial('vol', mu=y_hat, alpha=alpha, observed=total_quantity)
start = pm.find_MAP()
step = pm.NUTS()
# Inference
trace = pm.sample(3, start = start, step = step, cores =2, tune=0, chains=1) #I set it up this way for debugging purposes
```

Which results in the following:

```
logp = -inf, ||grad|| = 4,910.9: 100%|██████████| 3/3 [00:00<00:00, 386.14it/s]
Only 3 samples in chain.
Sequential sampling (1 chains in 1 job)
NUTS: [d, b, a, sigma_a, alpha]
0%| | 0/3 [00:00<?, ?it/s]
```

and subsequently, “ValueError: Bad initial energy: inf. The model might be misspecified.”

When I check the test_point, it is showing “vol” being the culprit for -inf

```
for RV in quantity_model_1.basic_RVs:
print(RV.name, RV.logp(quantity_model_1.test_point))
alpha_interval__ -1.3862943611198906
sigma_a_log__ -0.7698925914732455
a_log__ -1.539785182946491
b_log__ -5.389248140312718
d -1098.5111832542225
vol -inf
```

On further investigation:

```
y_hat.tag.test_value
array([1.83882346, 1.13897182, 1.60580621, ..., 1.39741757, 1.15156967,
1.45517467])
total_quantity
array([7., 2., 4., ..., 1., 1., 1.])
```

I have attached a pickle of the model. Use this to load it:

```
with open('quantity_model_fortest.pkl', 'rb') as buff:
data = pickle.load(buff)
quantity_model_1 = data['model']
```

Versions and main components

PyMC3 Version: 3.4.1

Theano Version: 1.0.2

Python Version: 3.6.5

Operating system: ubuntu

How did you install PyMC3: (conda/pip) pip

quantity_model_fortest.zip