test: Fix prerelease and pandas 3.0 test compat (#2457)

Author: TrevorBergeron

bigframes/core/compile/ibis_compiler/scalar_op_registry.py

@@ -169,6 +169,8 @@ def arctanh_op_impl(x: ibis_types.Value):
 @scalar_op_compiler.register_unary_op(ops.floor_op)
 def floor_op_impl(x: ibis_types.Value):
     x_numeric = typing.cast(ibis_types.NumericValue, x)
+    if x_numeric.type().is_boolean():
+        return x_numeric.cast(ibis_dtypes.Int64()).cast(ibis_dtypes.Float64())
     if x_numeric.type().is_integer():
         return x_numeric.cast(ibis_dtypes.Float64())
     if x_numeric.type().is_floating():
@@ -181,6 +183,8 @@ def floor_op_impl(x: ibis_types.Value):
 @scalar_op_compiler.register_unary_op(ops.ceil_op)
 def ceil_op_impl(x: ibis_types.Value):
     x_numeric = typing.cast(ibis_types.NumericValue, x)
+    if x_numeric.type().is_boolean():
+        return x_numeric.cast(ibis_dtypes.Int64()).cast(ibis_dtypes.Float64())
     if x_numeric.type().is_integer():
         return x_numeric.cast(ibis_dtypes.Float64())
     if x_numeric.type().is_floating():
@@ -1026,7 +1030,7 @@ def to_timedelta_op_impl(x: ibis_types.Value, op: ops.ToTimedeltaOp):
 
 @scalar_op_compiler.register_unary_op(ops.timedelta_floor_op)
 def timedelta_floor_op_impl(x: ibis_types.NumericValue):
-    return x.floor()
+    return ibis_api.case().when(x > ibis.literal(0), x.floor()).else_(x.ceil()).end()
 
 
 @scalar_op_compiler.register_unary_op(ops.RemoteFunctionOp, pass_op=True)

bigframes/core/compile/polars/lowering.py

@@ -174,12 +174,10 @@ def lower(self, expr: expression.OpExpression) -> expression.Expression:
             divisor.output_type
         ):
             # exact same as floordiv impl for timedelta
-            numeric_result = ops.floordiv_op.as_expr(
+            numeric_result = ops.div_op.as_expr(
                 ops.AsTypeOp(to_type=dtypes.INT_DTYPE).as_expr(dividend), divisor
             )
-            int_result = ops.AsTypeOp(to_type=dtypes.INT_DTYPE).as_expr(numeric_result)
-            return ops.AsTypeOp(to_type=dtypes.TIMEDELTA_DTYPE).as_expr(int_result)
-
+            return _numeric_to_timedelta(numeric_result)
         if (
             dividend.output_type == dtypes.BOOL_DTYPE
             and divisor.output_type == dtypes.BOOL_DTYPE
@@ -226,11 +224,10 @@ def lower(self, expr: expression.OpExpression) -> expression.Expression:
             divisor.output_type
         ):
             # this is pretty fragile as zero will break it, and must fit back into int
-            numeric_result = expr.op.as_expr(
+            numeric_result = ops.div_op.as_expr(
                 ops.AsTypeOp(to_type=dtypes.INT_DTYPE).as_expr(dividend), divisor
             )
-            int_result = ops.AsTypeOp(to_type=dtypes.INT_DTYPE).as_expr(numeric_result)
-            return ops.AsTypeOp(to_type=dtypes.TIMEDELTA_DTYPE).as_expr(int_result)
+            return _numeric_to_timedelta(numeric_result)
 
         if dividend.output_type == dtypes.BOOL_DTYPE:
             dividend = ops.AsTypeOp(to_type=dtypes.INT_DTYPE).as_expr(dividend)
@@ -319,6 +316,32 @@ def lower(self, expr: expression.OpExpression) -> expression.Expression:
         return expr
 
 
+class LowerCeilOp(op_lowering.OpLoweringRule):
+    @property
+    def op(self) -> type[ops.ScalarOp]:
+        return numeric_ops.CeilOp
+
+    def lower(self, expr: expression.OpExpression) -> expression.Expression:
+        assert isinstance(expr.op, numeric_ops.CeilOp)
+        arg = expr.children[0]
+        if arg.output_type in (dtypes.INT_DTYPE, dtypes.BOOL_DTYPE):
+            return expr.op.as_expr(ops.AsTypeOp(dtypes.FLOAT_DTYPE).as_expr(arg))
+        return expr
+
+
+class LowerFloorOp(op_lowering.OpLoweringRule):
+    @property
+    def op(self) -> type[ops.ScalarOp]:
+        return numeric_ops.FloorOp
+
+    def lower(self, expr: expression.OpExpression) -> expression.Expression:
+        assert isinstance(expr.op, numeric_ops.FloorOp)
+        arg = expr.children[0]
+        if arg.output_type in (dtypes.INT_DTYPE, dtypes.BOOL_DTYPE):
+            return expr.op.as_expr(ops.AsTypeOp(dtypes.FLOAT_DTYPE).as_expr(arg))
+        return expr
+
+
 class LowerIsinOp(op_lowering.OpLoweringRule):
     @property
     def op(self) -> type[ops.ScalarOp]:
@@ -465,8 +488,21 @@ def _lower_cast(cast_op: ops.AsTypeOp, arg: expression.Expression):
     LowerInvertOp(),
     LowerIsinOp(),
     LowerLenOp(),
+    LowerCeilOp(),
+    LowerFloorOp(),
 )
 
 
 def lower_ops_to_polars(root: bigframe_node.BigFrameNode) -> bigframe_node.BigFrameNode:
     return op_lowering.lower_ops(root, rules=POLARS_LOWERING_RULES)
+
+
+def _numeric_to_timedelta(expr: expression.Expression) -> expression.Expression:
+    """rounding logic used for emulating timedelta ops"""
+    rounded_value = ops.where_op.as_expr(
+        ops.floor_op.as_expr(expr),
+        ops.gt_op.as_expr(expr, expression.const(0)),
+        ops.ceil_op.as_expr(expr),
+    )
+    int_value = ops.AsTypeOp(to_type=dtypes.INT_DTYPE).as_expr(rounded_value)
+    return ops.AsTypeOp(to_type=dtypes.TIMEDELTA_DTYPE).as_expr(int_value)

bigframes/core/compile/sqlglot/expressions/common.py

@@ -0,0 +1,33 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import bigframes_vendored.sqlglot.expressions as sge
+
+
+def round_towards_zero(expr: sge.Expression):
+    """
+    Round a float value to to an integer, always rounding towards zero.
+
+    This is used to handle duration/timedelta emulation mostly.
+    """
+    return sge.Cast(
+        this=sge.If(
+            this=sge.GT(this=expr, expression=sge.convert(0)),
+            true=sge.Floor(this=expr),
+            false=sge.Ceil(this=expr),
+        ),
+        to="INT64",
+    )

bigframes/core/compile/sqlglot/expressions/numeric_ops.py

@@ -20,6 +20,7 @@
 from bigframes import dtypes
 from bigframes import operations as ops
 import bigframes.core.compile.sqlglot.expression_compiler as expression_compiler
+from bigframes.core.compile.sqlglot.expressions.common import round_towards_zero
 import bigframes.core.compile.sqlglot.expressions.constants as constants
 from bigframes.core.compile.sqlglot.expressions.typed_expr import TypedExpr
 from bigframes.operations import numeric_ops
@@ -467,7 +468,7 @@ def _(left: TypedExpr, right: TypedExpr) -> sge.Expression:
 
     result = sge.func("IEEE_DIVIDE", left_expr, right_expr)
     if left.dtype == dtypes.TIMEDELTA_DTYPE and dtypes.is_numeric(right.dtype):
-        return sge.Cast(this=sge.Floor(this=result), to="INT64")
+        return round_towards_zero(result)
     else:
         return result
 
@@ -510,7 +511,7 @@ def _(left: TypedExpr, right: TypedExpr) -> sge.Expression:
     )
 
     if dtypes.is_numeric(right.dtype) and left.dtype == dtypes.TIMEDELTA_DTYPE:
-        result = sge.Cast(this=sge.Floor(this=result), to="INT64")
+        result = round_towards_zero(sge.func("IEEE_DIVIDE", left_expr, right_expr))
 
     return result
 
@@ -578,7 +579,7 @@ def _(left: TypedExpr, right: TypedExpr) -> sge.Expression:
     if (dtypes.is_numeric(left.dtype) and right.dtype == dtypes.TIMEDELTA_DTYPE) or (
         left.dtype == dtypes.TIMEDELTA_DTYPE and dtypes.is_numeric(right.dtype)
     ):
-        return sge.Cast(this=sge.Floor(this=result), to="INT64")
+        return round_towards_zero(result)
     else:
         return result
 

bigframes/core/rewrite/timedeltas.py

@@ -206,12 +206,12 @@ def _rewrite_div_op(left: _TypedExpr, right: _TypedExpr) -> _TypedExpr:
 
 
 def _rewrite_floordiv_op(left: _TypedExpr, right: _TypedExpr) -> _TypedExpr:
-    result = _TypedExpr.create_op_expr(ops.floordiv_op, left, right)
-
     if left.dtype == dtypes.TIMEDELTA_DTYPE and dtypes.is_numeric(right.dtype):
-        return _TypedExpr.create_op_expr(ops.timedelta_floor_op, result)
+        return _TypedExpr.create_op_expr(
+            ops.timedelta_floor_op, _TypedExpr.create_op_expr(ops.div_op, left, right)
+        )
 
-    return result
+    return _TypedExpr.create_op_expr(ops.floordiv_op, left, right)
 
 
 def _rewrite_to_timedelta_op(op: ops.ToTimedeltaOp, arg: _TypedExpr):

bigframes/core/utils.py

@@ -113,13 +113,13 @@ def get_standardized_ids(
     """
     col_ids = [
         UNNAMED_COLUMN_ID
-        if col_label is None
+        if pd.isna(col_label)  # type: ignore
         else label_to_identifier(col_label, strict=strict)
         for col_label in col_labels
     ]
     idx_ids = [
         UNNAMED_INDEX_ID
-        if idx_label is None
+        if pd.isna(idx_label)  # type: ignore
         else label_to_identifier(idx_label, strict=strict)
         for idx_label in idx_labels
     ]

bigframes/dataframe.py

@@ -4398,7 +4398,7 @@ def to_excel(
         **kwargs,
     ) -> None:
         return self.to_pandas(allow_large_results=allow_large_results).to_excel(
-            excel_writer, sheet_name, **kwargs
+            excel_writer, sheet_name=sheet_name, **kwargs
         )
 
     def to_latex(

bigframes/ml/metrics/_metrics.py

@@ -214,7 +214,7 @@ def confusion_matrix(
         y_true = row["y_true"]
         y_pred = row["y_pred"]
         count = row["dummy"]
-        confusion_matrix[y_pred][y_true] = count
+        confusion_matrix.at[y_true, y_pred] = count
 
     return confusion_matrix
 
@@ -251,7 +251,7 @@ def recall_score(
         / is_accurate.groupby(y_true_series).count()
     ).to_pandas()
 
-    recall_score = pd.Series(0, index=index)
+    recall_score = pd.Series(0.0, index=index)
     for i in recall_score.index:
         recall_score.loc[i] = recall.loc[i]
 
@@ -321,7 +321,7 @@ def _precision_score_per_label(y_true: bpd.Series, y_pred: bpd.Series) -> pd.Ser
         is_accurate.groupby(y_pred).sum() / is_accurate.groupby(y_pred).count()
     ).to_pandas()
 
-    precision_score = pd.Series(0, index=index)
+    precision_score = pd.Series(0.0, index=index)
     for i in precision.index:
         precision_score.loc[i] = precision.loc[i]
 
@@ -366,7 +366,7 @@ def f1_score(
     recall = recall_score(y_true_series, y_pred_series, average=None)
     precision = precision_score(y_true_series, y_pred_series, average=None)
 
-    f1_score = pd.Series(0, index=recall.index)
+    f1_score = pd.Series(0.0, index=recall.index)
     for index in recall.index:
         if precision[index] + recall[index] != 0:
             f1_score[index] = (

bigframes/series.py

@@ -2292,7 +2292,7 @@ def to_excel(
         self, excel_writer, sheet_name="Sheet1", *, allow_large_results=None, **kwargs
     ) -> None:
         return self.to_pandas(allow_large_results=allow_large_results).to_excel(
-            excel_writer, sheet_name, **kwargs
+            excel_writer, sheet_name=sheet_name, **kwargs
         )
 
     def to_json(

bigframes/session/polars_executor.py

@@ -79,6 +79,8 @@
     numeric_ops.SubOp,
     numeric_ops.MulOp,
     numeric_ops.DivOp,
+    numeric_ops.CeilOp,
+    numeric_ops.FloorOp,
     numeric_ops.FloorDivOp,
     numeric_ops.ModOp,
     generic_ops.AsTypeOp,