add option for fitting to the whole view instead of wrk set

Author: antheas

src/pasteur/kedro/pipelines/main.py

@@ -124,8 +124,18 @@ def generate_pipelines(
         )
 
         # Create view transform pipeline that can run as part of ingest
+        if view.fit_global:
+            pipe_fit = create_fit_pipeline(view, all_types, modules, 'view')
+            + create_transform_pipeline(
+                view,
+                'view',
+                all_types,
+            )
+        else:
+            pipe_fit = create_fit_pipeline(view, all_types, modules, wrk_split)
+
         pipe_transform = (
-            create_fit_pipeline(view, all_types, modules, wrk_split)
+            pipe_fit
             + create_transform_pipeline(
                 view,
                 wrk_split,

src/pasteur/view.py

@@ -167,6 +167,12 @@ class View(Module):
     """
     trn_deps: dict[str, list[str]] = {}
     parameters: dict[str, Any] | str | None = None
+    
+    """ If true, transformers and encoders for this view will be fit on the global
+    dataset. Resolves encoding errors that stem from sampling the partial view.
+    When true, subsampling the view is not possible during transformation and 
+    encoding, which may add significant overhead."""
+    fit_global: bool = False
 
     def __init__(self, **_) -> None:
         pass