WIP

formulaic/materializers/base.py

@@ -522,7 +522,7 @@ def _encode_evaled_factor(
         drop_rows: set,
         reduced_rank: bool = False,
     ) -> Dict[str, Any]:
-        if not isinstance(factor.values, dict) or not factor.metadata.encoded:
+        if not factor.metadata.encoded:
             if factor.expr in self.encoded_cache:
                 encoded = self.encoded_cache[factor.expr]
             elif (factor.expr, reduced_rank) in self.encoded_cache:
@@ -552,7 +552,9 @@ def wrapped(values, metadata, state, *args, **kwargs):
                                     if nested_state:
                                         state[k] = nested_state
                             if isinstance(values, FactorValues):
-                                return FactorValues(encoded, metadata=values.__formulaic_metadata__)
+                                return FactorValues(
+                                    encoded, metadata=values.__formulaic_metadata__
+                                )
                             return encoded
                         return f(values, metadata, state, *args, **kwargs)
 
@@ -618,7 +620,7 @@ def wrapped(values, metadata, state, *args, **kwargs):
 
                 self.encoded_cache[cache_key] = encoded
         else:
-            encoded = factor.values
+            encoded = as_columns(factor.values)
 
         encoded = FactorValues(
             encoded,

formulaic/materializers/pandas.py

@@ -6,6 +6,9 @@
 import pandas
 import scipy.sparse as spsparse
 from interface_meta import override
+from formulaic.materializers.types.factor_values import FactorValues
+
+from formulaic.utils.cast import as_columns
 
 from .base import FormulaMaterializer
 from .types import NAAction
@@ -74,17 +77,21 @@ def _encode_categorical(
         # Even though we could reduce rank here, we do not, so that the same
         # encoding can be cached for both reduced and unreduced rank. The
         # rank will be reduced in the _encode_evaled_factor method.
-        from formulaic.transforms import contrasts
+        from formulaic.transforms import encode_contrasts
 
         if drop_rows:
             values = values.drop(index=values.index[drop_rows])
-        return contrasts(
+        encoded = encode_contrasts(
             values,
             reduced_rank=False,
             _metadata=metadata,
             _state=encoder_state,
             _spec=spec,
         )
+        return FactorValues(
+            as_columns(encoded),
+            metadata=encoded.__formulaic_metadata__,
+        )
 
     @override
     def _get_columns_for_term(self, factors, spec, scale=1):
@@ -132,7 +139,6 @@ def _get_columns_for_term(self, factors, spec, scale=1):
 
     @override
     def _combine_columns(self, cols, spec, drop_rows):
-
         # If we are outputing a pandas DataFrame, explicitly override index
         # in case transforms/etc have lost track of it.
         if spec.output == "pandas":

formulaic/materializers/types/factor_values.py

@@ -72,7 +72,9 @@ def __init__(
         drop_field: Optional[str] = MISSING,
         format: str = MISSING,
         encoded: bool = MISSING,
-        encoder: Optional[Callable[[Any, bool, List[int], Dict[str, Any]], Any]] = MISSING,
+        encoder: Optional[
+            Callable[[Any, bool, List[int], Dict[str, Any]], Any]
+        ] = MISSING,
     ):
         metadata_constructor = FactorValuesMetadata
         metadata_kwargs = dict(

formulaic/transforms/__init__.py

@@ -2,10 +2,21 @@
 
 from .basis_spline import basis_spline
 from .identity import identity
-from .contrasts import contrasts
+from .contrasts import C, encode_contrasts, ContrastsRegistry
 from .poly import poly
 from .scale import center, scale
 
+__all__ = [
+    "basis_spline",
+    "identity",
+    "C",
+    "encode_contrasts",
+    "ContrastsRegistry",
+    "poly",
+    "center",
+    "scale",
+    "TRANSFORMS",
+]
 
 TRANSFORMS = {
     # Common transforms
@@ -21,6 +32,7 @@
     "center": center,
     "poly": poly,
     "scale": scale,
-    "C": contrasts,
+    "C": C,
+    "contr": ContrastsRegistry,
     "I": identity,
 }