Add a way to extract observations from pymc3 trace #219
Conversation
arviz/utils/xarray_utils.py
Outdated
| @@ -277,6 +336,7 @@ def to_inference_data(self): | |||
| 'sample_stats': self.sample_stats_to_xarray(), | |||
| 'posterior_predictive': self.posterior_predictive_to_xarray(), | |||
| 'prior': self.prior_to_xarray(), | |||
| 'data': self.data_to_xarray(), | |||
There was a problem hiding this comment.
The only thing that concerns me about this change is that we use the variable name data throughout the library to refer to posterior.
Would you alright if I opened an issue and/or made a pull request to rename this in the various plots? For example traceplot
https://github.com/arviz-devs/arviz/blob/master/arviz/plots/traceplot.py
There was a problem hiding this comment.
I don't like it either. data is really arbitrary thing.
Could it be observed, observed_data or something else?
There was a problem hiding this comment.
I think calling it data was a nod to stan's data section :)
I updated it to 'observed_data', and also updated schema.md. @canyon289 let me know if that's ok with you, too.
|
Also, should we have We can deal with prior later? |
|
Merging this, but will create an issue for how/whether to store other data associated with a model. |
Note also that I refactored the
numpy_to_xarrayfunction so that consistent names would be generated. Now theposterior_predictiveanddatavariables will have the same default dimension names.This change will make it easier to add and test the
ppc_plot.