what is the recommended workflow to do posterior predictive sample

[hyang336]

Hi, I am trying to figure out the best way to do posterior predictive sample in HSSM. The HSSM class has a method "sample_posterior_predictive", but when I tried to use that with an arviz InferenceData, it returned an error:"

model=hssm.HSSM(data=data)
model.sample_posterior_predictive(idata=idata)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/hssm/hssm.py", line 715, in sample_posterior_predictive
    self.model.predict(
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/bambi/models.py", line 832, in predict
    means_dict[var_name] = component.predict(
                           ^^^^^^^^^^^^^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/bambi/model_components.py", line 185, in predict
    linear_predictor += self.predict_common(
                        ^^^^^^^^^^^^^^^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/bambi/model_components.py", line 300, in predict_common
    linear_predictor += xr.dot(X, b)
                        ^^^^^^^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/util/deprecation_helpers.py", line 140, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/computation.py", line 1913, in dot
    result = apply_ufunc(
             ^^^^^^^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/computation.py", line 1266, in apply_ufunc
    return apply_dataarray_vfunc(
           ^^^^^^^^^^^^^^^^^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/computation.py", line 291, in apply_dataarray_vfunc
    deep_align(
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/alignment.py", line 946, in deep_align
    aligned = align(
              ^^^^^^
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/alignment.py", line 882, in align
    aligner.align()
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/alignment.py", line 575, in align
    self.assert_unindexed_dim_sizes_equal()
  File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/xarray/core/alignment.py", line 476, in assert_unindexed_dim_sizes_equal
    raise ValueError(
ValueError: cannot reindex or align along dimension '__variables__' because of conflicting dimension sizes: {1, 2692} (note: an index is found along that dimension with size=2692)

Also is there a way to omit the data variable in the HSSM initialization, given that it is not used for posterior predictive sampling anyway?

[digicosmos86]

Hi,

You will need to have the original model with which you sampled the idata object. You don't need to sample the model again. You can just build the model with the parameter specifications and then do model._inferenced_obj=idata. Then you can do model.sample_posterior_predictive()

We will soon add a attach_traces() method to make this look less hacky

[hyang336]

Hi,

Thanks for the reply. Do I need to specify the exact same priors as well, or just the same data and model type?

[digicosmos86]

You will need to specify the exact same model

[hyang336]

It returned an error
ValueError: The model has not been sampled yet. Please either provide an idata object or sample the model first.

Should I pass the idata to sample_posterior_predictive() instead?

[digicosmos86]

Sorry for the typo. It's model._inference_obj that should be assigned the idata object