craft#

craft(spec)[source]#

Instantiate an object from the specification string.

The craft utility can be used to deserialize an estimator specification string, including composites such as pipelines.

It takes a specification string and returns an sktime estimator, class, or object, corresponding to that string.

Specification strings can be:

  • simple expressions such as "NaiveForecaster" or "NaiveForecaster(sp=2)"

  • compositions such as "Deseasonalizer() * NaiveForecaster()"

  • a longer block of code, closing with a return statement, e.g., the string block

deseason = Deseasonalizer()
naive = NaiveForecaster()
return deseason * naive

The craft utility is useful as a serialization / deserialization pair, together with str coercion (or commandline printing) of an unfitted estimator.

craft recognizes estimators present in sktime and scikit-learn, and base python.

Parameters:
specstr, sktime/skbase compatible object specification

i.e., a string that executes to construct an object if all imports were present imports inferred are of any classes in the scope of all_estimators

  • option 1: a string that evaluates to an estimator

  • option 2: a sequence of assignments in valid python code, with the object to be defined preceded by a “return”. assignments can use names of classes as if all imports were present

Returns:
objskbase BaseObject descendant, constructed from spec

this will have the property that spec == str(obj) (up to formatting)

Examples

>>> from sktime.registry import craft

Example 1: simple estimator

  • serialized as the string "NaiveForecaster(sp=2)"

  • deserialized as the estimator NaiveForecaster(sp=2)

>>> spec = "NaiveForecaster(sp=2)"
>>> est = craft(spec)
>>> print(est)
NaiveForecaster(sp=2)

Example 2: composite estimator

  • serialized as the string "Deseasonalizer() * NaiveForecaster()"

  • deserialized as the estimator Deseasonalizer() * NaiveForecaster(), same as ForecastingPipeline([Deseasonalizer(), NaiveForecaster()])

>>> spec = "Deseasonalizer() * NaiveForecaster()"
>>> est = craft(spec)

Example 3: longer code block

  • serialized as a code block with assignments and return statement

  • deserialized as the estimator defined in the return statement

>>> spec = '''
... deseason = Deseasonalizer()
... naive = NaiveForecaster()
... return deseason * naive
... '''
>>> est = craft(spec)
On this page