Shortcuts

torcharrow.Column.map

Column.map(arg: ty.Union[ty.Dict, ty.Callable], na_action: ty.Literal['ignore', None] = None, dtype: ty.Optional[dt.DType] = None, columns: ty.Optional[ty.List[str]] = None)

Maps rows according to input correspondence.

Parameters:
  • callable (arg - dict or) – If arg is a dict then input is mapped using this dict and non-mapped values become null. If arg is a callable, this is treated as a user-defined function (UDF) which is invoked on each element of the input. Callables must be global functions or methods on class instances, lambdas are not supported.

  • None (default) – If your UDF returns null for null input, selecting “ignore” is an efficiency improvement where map will avoid calling your UDF on null values. If None, aways calls the UDF.

  • None – If your UDF returns null for null input, selecting “ignore” is an efficiency improvement where map will avoid calling your UDF on null values. If None, aways calls the UDF.

  • DType (dtype -) – DType is used to force the output type. DType is required if result type != item type.

  • None – DType is used to force the output type. DType is required if result type != item type.

  • names (columns - list of column) – Determines which columns to provide to the mapping dict or UDF.

  • None – Determines which columns to provide to the mapping dict or UDF.

See also

flatmap, filter

Examples

>>> import torcharrow as ta
>>> ta.column([1,2,None,4]).map({1:111})
0  111
1  None
2  None
3  None
dtype: Int64(nullable=True), length: 4, null_count: 3

Using a defaultdict to provide a missing value:

>>> from collections import defaultdict
>>> ta.column([1,2,None,4]).map(defaultdict(lambda: -1, {1:111}))
0  111
1   -1
2   -1
3   -1
dtype: Int64(nullable=True), length: 4, null_count: 0

Using user-supplied python function:

>>> def add_ten(num):
>>>     return num + 10
>>>
>>> ta.column([1,2,None,4]).map(add_ten, na_action='ignore')
0  11
1  12
2  None
3  14
dtype: Int64(nullable=True), length: 4, null_count: 1

Note that .map(add_ten, na_action=None) in the example above would fail with a type error since addten is not defined for None/null. To pass nulls to a UDF, the UDF needs to prepare for it:

>>> def add_ten_or_0(num):
>>>     return 0 if num is None else num + 10
>>>
>>> ta.column([1,2,None,4]).map(add_ten_or_0, na_action=None)
0  11
1  12
2   0
3  14
dtype: Int64(nullable=True), length: 4, null_count: 0

Mapping to different types requires a dtype parameter:

>>> ta.column([1,2,None,4]).map(str, dtype=dt.string)
0  '1'
1  '2'
2  'None'
3  '4'
dtype: string, length: 4, null_count: 0

Mapping over a DataFrame, the UDF gets the whole row as a tuple:

>>> def add_unary(tup):
>>>     return tup[0]+tup[1]
>>>
>>> ta.dataframe({'a': [1,2,3], 'b': [1,2,3]}).map(add_unary , dtype = dt.int64)
0  2
1  4
2  6
dtype: int64, length: 3, null_count: 0

Multi-parameter UDFs:

>>> def add_binary(a,b):
>>>     return a + b
>>>
>>> ta.dataframe({'a': [1,2,3], 'b': ['a', 'b', 'c'], 'c':[1,2,3]}).map(add_binary, columns = ['a','c'], dtype = dt.int64)
0  2
1  4
2  6
dtype: int64, length: 3, null_count: 0

Multi-return UDFs - functions that return more than one column can be specified by returning a DataFrame (also known as struct column); providing the return dtype is mandatory:

>>> ta.dataframe({'a': [17, 29, 30], 'b': [3,5,11]}).map(divmod, columns= ['a','b'], dtype = dt.Struct([dt.Field('quotient', dt.int64), dt.Field('remainder', dt.int64)]))
  index    quotient    remainder
-------  ----------  -----------
      0           5            2
      1           5            4
      2           2            8
dtype: Struct([Field('quotient', int64), Field('remainder', int64)]), count: 3, null_count: 0

UDFs with state can be written by capturing the state in a (data)class and use a method as a delegate:

>>> def fib(n):
>>>     if n == 0:
>>>         return 0
>>>     elif n == 1 or n == 2:
>>>         return 1
>>>     else:
>>>         return fib(n-1) + fib(n-2)
>>>
>>> from dataclasses import dataclass
>>> @dataclass
>>> class State:
>>>     state: int
>>>     def __post_init__(self):
>>>         self.state = fib(self.state)
>>>     def add_fib(self, x):
>>>         return self.state+x
>>>
>>> m = State(10)
>>> ta.column([1,2,3]).map(m.add_fib)
0  56
1  57
2  58
dtype: int64, length: 3, null_count: 0

Docs

Access comprehensive developer documentation for PyTorch

View Docs

Tutorials

Get in-depth tutorials for beginners and advanced developers

View Tutorials

Resources

Find development resources and get your questions answered

View Resources