fix(layer): cast polars decimal columns to float

plotnine/layer.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import decimal
 import typing
 from copy import copy, deepcopy
 from typing import Iterable, List, cast, overload
@@ -257,6 +258,8 @@ def _make_layer_data(self, plot_data: DataLike | None):
             else:
                 raise TypeError(f"Data has a bad type: {type(self.data)}")
 
+        self.data = _decimal_columns_to_float(self.data)
+
     def _make_layer_mapping(self, plot_mapping: aes):
         """
         Create the aesthetic mappings to be used by this layer
@@ -778,3 +781,27 @@ def _resolve_position(
         raise PlotnineError(f"Unknown position of type {type(position_spec)}")
 
     return klass()
+
+
+def _decimal_columns_to_float(data: pd.DataFrame) -> pd.DataFrame:
+    """
+    Cast columns of decimal.Decimal values to float
+
+    polars' `to_pandas()` maps a Decimal column onto a pandas object column
+    holding `decimal.Decimal` values. plotnine treats object columns as
+    discrete, so such a column cannot be used with a continuous scale. Casting
+    it to float lets it behave like any other numeric column.
+
+    This runs for every dataframe, so we avoid the expense of `.dropna()`
+    (which builds a filtered copy) and use a mask to peek at the first
+    non-null value instead.
+    """
+    for col in data.columns:
+        series = data[col]
+        if series.dtype == object:
+            mask = series.notna()
+            if mask.any() and isinstance(
+                series[mask.idxmax()], decimal.Decimal
+            ):
+                data[col] = series.astype("float64")
+    return data

tests/test_ggplot_internals.py

@@ -12,6 +12,7 @@
     coord_trans,
     facet_null,
     geom_bar,
+    geom_col,
     geom_histogram,
     geom_line,
     geom_point,
@@ -21,6 +22,7 @@
     labs,
     lims,
     scale_x_continuous,
+    scale_y_continuous,
     stage,
     stat_identity,
     theme,
@@ -365,6 +367,26 @@ def to_pandas(self):
     assert p2 == "to_pandas"
 
 
+def test_decimal_columns():
+    # A pandas object column of decimal.Decimal values (e.g. from polars'
+    # to_pandas()) would otherwise be treated as discrete and rejected by a
+    # continuous scale. See GH #1033.
+    from decimal import Decimal
+
+    data = pd.DataFrame(
+        {
+            "x": ["a", "b", "c"],
+            "y": [Decimal(1), Decimal(2), Decimal(3)],
+        }
+    )
+    p = (
+        ggplot(data, aes("x", "y"))
+        + geom_col()
+        + scale_y_continuous(limits=(0, 5))
+    )
+    p._build()
+
+
 def test_callable_as_data():
     def _fn(data):
         return data.rename(columns={"xx": "x", "yy": "y"})