deferred

skrub.deferred(func)

Wrap function calls in a DataOp DataOp.

When this decorator is applied, the resulting function returns DataOps. The returned DataOp wraps the call to the original function, and the call is executed when the DataOp is evaluated. This allows including calls to any function as a step in a learner, rather than executing it immediately.

See the examples gallery for an in-depth explanation of skrub DataOps and deferred.

Parameters:

funcfunction

The function to wrap

Returns:

A new function

When called, rather than applying the original function immediately, it returns a DataOp. Evaluating the DataOp applies the original function.

See also

as_data_op

Create a DataOp that evaluates to the given value.

DataOp

Representation of a computation that can be used to build ML estimators.

Examples

>>> def tokenize(text):
...     words = text.split()
...     return [w for w in words if w not in ['the', 'of']]
>>> tokenize('the first day of the week')
['first', 'day', 'week']

>>> import skrub
>>> text = skrub.var('text')

Calling tokenize on a skrub DataOp raises an exception: tokenize tries to iterate immediately over the tokens to remove stop words, but the text will only be known when we run the learner.

>>> tokens = tokenize(text)
Traceback (most recent call last):
    ...
TypeError: This object is a DataOp that will be evaluated later, when your learner runs. So it is not possible to eagerly iterate over it now.

We can defer the call to tokenize until we are evaluating the DataOp:

>>> tokens = skrub.deferred(tokenize)(text)
>>> tokens
<Call 'tokenize'>
>>> tokens.skb.eval({'text': 'the first month of the year'})
['first', 'month', 'year']

Like any decorator deferred can be called explicitly as shown above or used with the @ syntax:

>>> @skrub.deferred
... def log(x):
...     print('INFO x =', x)
...     return x
>>> x = skrub.var('x')
>>> e = log(x)
>>> e.skb.eval({'x': 3})
INFO x = 3
3