# Issues when trying to do out of sample prediction

**URL:** <https://discourse.pymc.io/t/issues-when-trying-to-do-out-of-sample-prediction/14216>\
**Category:** v5\
**Created:** [April 9, 2024, 1:53pm UTC](https://discourse.pymc.io/t/issues-when-trying-to-do-out-of-sample-prediction/14216 "2024-04-09T13:53:01Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![luca.pamparana](https://avatars.discourse-cdn.com/v4/letter/l/b9bd4f/32.png) [@luca.pamparana](https://discourse.pymc.io/u/luca.pamparana)\
**Post date:** [April 9, 2024, 1:53pm UTC](https://discourse.pymc.io/t/issues-when-trying-to-do-out-of-sample-prediction/14216/1 "2024-04-09T13:53:01Z")

</div>

I am using PYMC version `5.12.0` on python `3.11.8`.  
I am trying to do out of sample prediction with my model and I tried to follow what have been done in the examples. My model is defined as follows:

```auto
with pm.Model() as hierarchical_model:
    # Hyperpriors
    stop_effect_mean = pm.Normal('stop_effect_mean', mu=0., sigma=10.)
    time_effect_mean = pm.Normal('time_effect_mean', mu=0., sigma=10.)
    day_type_effect_mean = pm.Normal('day_type_effect_mean', mu=0., sigma=10.)
    
    # Individual level effects
    stop_effect = pm.Normal('stop_effect', mu=stop_effect_mean, sigma=1., shape=n_stops)
    time_effect = pm.Normal('time_effect', mu=time_effect_mean, sigma=1., shape=n_times)
    day_type_effect = pm.Normal('day_type_effect', mu=day_type_effect_mean, sigma=1., shape=n_day_types)
    
    bus_stops_idx_shared = pm.MutableData('bus_stops_idx', bus_stops_idx)
    times_of_day_idx_shared = pm.MutableData('times_of_day_idx', times_of_day_idx)
    day_types_idx_shared = pm.MutableData('day_types_idx', day_types_idx)
    
    # Model expectation: log-link function
    log_lambda = (stop_effect[bus_stops_idx_shared] +
                  time_effect[times_of_day_idx_shared] +
                  day_type_effect[day_types_idx_shared])
    
    # Poisson Likelihood
    observed_counts = pm.Poisson('observed_counts', mu=pm.math.exp(log_lambda), observed=counts)

    # Sampling
    trace = pm.sample()

```

I can train this model fine and then I am trying to do prediction as follows:

```auto
with hierarchical_model:
    # Update the shared variables with new test data indices
    pm.set_data({
        'bus_stops_idx': bus_stops_idx_test,  
        'times_of_day_idx': times_of_day_idx_test,
        'day_types_idx': day_types_idx_test,
    })
    
    # Generate posterior predictive samples for the test data
    ppc = pm.sample_posterior_predictive(trace, predictions=True)

```

However, this results in:

```auto
ValueError: shape mismatch: objects cannot be broadcast to a single shape. Mismatch is between arg 0 with shape (6720,) and arg 1 with shape (3360,).

```

This seems that it is expecting the shape to be of the training data that I had used. However, i thought using set\_data would allow me to do this. I am at a loss on what to try here.

---

<div class="post-metadata">

**Author:** ![ricardoV94](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/ricardov94/32/5775_2.png) [@ricardoV94](https://discourse.pymc.io/u/ricardoV94)\
**Post date:** [April 9, 2024, 2:02pm UTC](https://discourse.pymc.io/t/issues-when-trying-to-do-out-of-sample-prediction/14216/2 "2024-04-09T14:02:54Z")

</div>

Try setting `shape=mu.shape` in your observed variable. It’s shown in the documentation of `set_data`: [pymc.set\_data — PyMC v5.7.1 documentation](https://www.pymc.io/projects/docs/en/v5.7.1/api/generated/pymc.set_data.html)

---

<div class="post-metadata">

**Author:** ![luca.pamparana](https://avatars.discourse-cdn.com/v4/letter/l/b9bd4f/32.png) [@luca.pamparana](https://discourse.pymc.io/u/luca.pamparana)\
**Post date:** [April 9, 2024, 2:46pm UTC](https://discourse.pymc.io/t/issues-when-trying-to-do-out-of-sample-prediction/14216/3 "2024-04-09T14:46:44Z")

</div>

That did the trick! Thank you!