# How to make out of sample predictions with categorical variables? Implementation question from Rethinking 2 Book

**URL:** https://discourse.pymc.io/t/how-to-make-out-of-sample-predictions-with-categorical-variables-implementation-question-from-rethinking-2-book/5502
**Category:** Questions
**Created:** [July 22, 2020, 8:25pm UTC](https://discourse.pymc.io/t/how-to-make-out-of-sample-predictions-with-categorical-variables-implementation-question-from-rethinking-2-book/5502 "2020-07-22T20:25:41Z")
**Posts on this page:** 2
**Page:** 1

<div class="post-metadata">

### Author: ![mldl920](https://avatars.discourse-cdn.com/v4/letter/m/7cd45c/32.png) [@mldl920](https://discourse.pymc.io/u/mldl920)
#### Post date: [July 22, 2020, 8:25pm UTC](https://discourse.pymc.io/t/how-to-make-out-of-sample-predictions-with-categorical-variables-implementation-question-from-rethinking-2-book/5502/1 "2020-07-22T20:25:41Z")

</div>

Hi how’s it going? I’m going through Rethinking 2, and was wondering if you can help me finish this code in terms of making out of sample predictions.

Here is my simple example of what I have so far…

```auto
data = pd.DataFrame({'var1':np.random.random_sample(12),'cat_var':['a','b']*6,'y':[np.random.randint(5,20) for val in range(12)]})

```

![Screenshot from 2020-07-22 16-27-48](https://canada1.discourse-cdn.com/flex036/uploads/pymc3/original/2X/d/d97beaed6afdbc5db907443cf0430d387557796e.png)

```auto
data = pd.get_dummies(data,columns=['cat_var'])
cid = pd.Categorical(data["cat_var"])

with pm.Model() as m_5_1:
    a = pm.Normal("a", 1, 0.1, shape=cid.categories.size)
    b = pm.Normal("b", 0, 0.3)
    sigma = pm.Uniform("sigma", 0,1)
    mu = pm.Deterministic("mu", a[cid] + b * data['var1'])
    y = pm.Normal(
        "y",mu=mu, sigma=sigma, observed=y.values
    )
    trace = pm.sample()

```

Now if I were to just make out of sample predictions without the categorical variable, I would do:

```auto
new_data_0 = xr.DataArray(
    newdata['var1'],
    dims=["pred_id"]
)

pred_mean = (
    trace["a"][:newdata.shape[0]] +
    trace["b"][:newdata.shape[0]] * new_data_0 
)

rng = np.random.default_rng()
predictions = xr.apply_ufunc(lambda mu, sd: rng.normal(mu, sd), pred_mean, trace["sigma"][:newdata.shape[0]])

```

How would I add the categorical variable from new unseen data into this xr datatype format for predictions? In this case 0 for the letter “a”" and 1 for the letter “b”.

Thank you!

---

<div class="post-metadata">

### Author: ![AlexAndorra](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/alexandorra/32/9142_2.png) [@AlexAndorra](https://discourse.pymc.io/u/AlexAndorra)
#### Post date: [July 23, 2020, 9:10am UTC](https://discourse.pymc.io/t/how-to-make-out-of-sample-predictions-with-categorical-variables-implementation-question-from-rethinking-2-book/5502/2 "2020-07-23T09:10:34Z")

</div>

Hi,  
I think you can use a combination of `pm.Data` and `pm.sample` to do that – more PyMC-idiomatic.  
The repo of the [port of Rethinking\_2 to PyMC](https://github.com/pymc-devs/resources/tree/master/Rethinking_2) should have all the help you need 😉  
PyMCheers 🖖