[BUG] Series UDFs documentation bug

[sarahyurick]

Hi, I was going through the Overview of User Defined Functions with cuDF documentation and tried out the first example, but got this error:

import numpy as np

import cudf
from cudf.datasets import randomdata

df = randomdata(nrows=10, dtypes={'a':float, 'b':bool, 'c':str}, seed=12)
# df.head()

def udf(x):
    if x > 0:
        return x + 5
    else:
        return x - 5

df['a'].applymap(udf)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
/tmp/ipykernel_36751/1762304450.py in <module>
----> 1 df['a'].applymap(udf)

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/series.py in applymap(self, udf, out_dtype)
   3904         if not callable(udf):
   3905             raise ValueError("Input UDF must be a callable object.")
-> 3906         return self._from_data({self.name: self._unaryop(udf)}, self._index)
   3907 
   3908     #

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/frame.py in _unaryop(self, op)
   3422     def _unaryop(self, op):
   3423         data_columns = (col.unary_operator(op) for col in self._columns)
-> 3424         return self.__class__._from_data(
   3425             zip(self._column_names, data_columns), self._index
   3426         )

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/series.py in _from_data(cls, data, index, name)
    278         Construct the Series from a ColumnAccessor
    279         """
--> 280         out: Series = super()._from_data(data, index, name)
    281         if index is None:
    282             out._index = RangeIndex(out._data.nrows)

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/frame.py in _from_data(cls, data, index, name)
   4449     ):
   4450 
-> 4451         out = super()._from_data(data, index)
   4452         if name is not None:
   4453             out.name = name

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/frame.py in _from_data(cls, data, index)
     71     ):
     72         obj = cls.__new__(cls)
---> 73         libcudf.table.Table.__init__(obj, data, index)
     74         return obj
     75 

cudf/_lib/table.pyx in cudf._lib.table.Table.__init__()

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/column_accessor.py in __init__(self, data, multiindex, level_names)
    119             self._data = {}
    120             if data:
--> 121                 data = dict(data)
    122                 # Faster than next(iter(data.values()))
    123                 column_length = len(data[next(iter(data))])

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/frame.py in <genexpr>(.0)
   3421 
   3422     def _unaryop(self, op):
-> 3423         data_columns = (col.unary_operator(op) for col in self._columns)
   3424         return self.__class__._from_data(
   3425             zip(self._column_names, data_columns), self._index

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/column/numerical.py in unary_operator(self, unaryop)
    120 
    121     def unary_operator(self, unaryop: str) -> ColumnBase:
--> 122         return _numeric_column_unaryop(self, op=unaryop)
    123 
    124     def binary_operator(

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/core/column/numerical.py in _numeric_column_unaryop(operand, op)
    592 def _numeric_column_unaryop(operand: ColumnBase, op: str) -> ColumnBase:
    593     if callable(op):
--> 594         return libcudf.transform.transform(operand, op)
    595 
    596     op = libcudf.unary.UnaryOp[op.upper()]

cudf/_lib/transform.pyx in cudf._lib.transform.transform()

~/miniconda3/envs/cudf_dev/lib/python3.8/site-packages/cudf/utils/cudautils.py in compile_udf(udf, type_signature)
    261         udf, type_signature, device=True
    262     )
--> 263     if not isinstance(return_type, cudf.core.udf.typing.MaskedType):
    264         output_type = numpy_support.as_dtype(return_type).type
    265     else:

AttributeError: module 'cudf.core' has no attribute 'udf'

[beckernick]

This may have shaken out during the ongoing python refactor. We still have a cudf.core.udf namespace, but the __init__.py in cudf.core is now empty

https://github.com/rapidsai/cudf/blob/58692648043eb07d92e8462224c3eee9d8febb3d/python/cudf/cudf/core/__init__.py

[beckernick]

Looks like it was removed in #8949

[efajardo-nv]

hi @beckernick. fyi we just started getting the same error in the clx 21.10 nightly builds where we use udf's.

[shwina]

fix coming up

[efajardo-nv]

thanks ashwin!