as_data_op

skrub.as_data_op(value)

Create a DataOp DataOp that evaluates to the given value.

This wraps any object in a DataOp. When the DataOp is evaluated, the result is the provided value. This has a similar role as deferred(), but for any object rather than for functions.

Parameters:

valueobject

The result of evaluating the DataOp

Returns:

a DataOp that evaluates to the given value

See also

deferred

Wrap function calls in a DataOp.

DataOp

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

Examples

>>> import skrub
>>> data_source = skrub.var('source')
>>> data_path = skrub.as_data_op(
...     {"local": "data.parquet", "remote": "remote/data.parquet"}
... )[data_source]
>>> data_path.skb.eval({'source': 'remote'})
'remote/data.parquet'

Turning the dictionary into a DataOp defers the lookup of data_source until it has been evaluated when the learner runs.

The example above is somewhat contrived, but as_data_op is often useful with choices.

>>> x1 = skrub.var('x1')
>>> x2 = skrub.var('x2')
>>> features = skrub.choose_from({'x1': x1, 'x2': x2}, name='features')
>>> skrub.as_data_op(features).skb.apply(skrub.TableVectorizer())
<Apply TableVectorizer>

In fact, this can even be shortened slightly by using the choice’s method as_data_op:

>>> features.as_data_op().skb.apply(skrub.TableVectorizer())
<Apply TableVectorizer>