The PDF model class should not be computing also predictions.
The idea is to refactor the code such that pred and pdf are not tight together anymore.
For example, in the likelihood.py module instead of having
self.pred_and_pdf = pdf_model.pred_and_pdf_func(
fit_xgrid, forward_map=forward_map
)
We should have something like
self.pdf_func = pdf.grid_values_func
self.pred = forward_map
where forward_map is to be understood as a node of colibri.
To achieve this one can simply rename the make_pred_data node to forward_map.
The user that is developing its own model can then override forward_map with its own ideas.
The PDF model class should not be computing also predictions.
The idea is to refactor the code such that pred and pdf are not tight together anymore.
For example, in the
likelihood.pymodule instead of havingWe should have something like
where
forward_mapis to be understood as a node of colibri.To achieve this one can simply rename the
make_pred_datanode toforward_map.The user that is developing its own model can then override
forward_mapwith its own ideas.