Skip to content

flowmachine.features.utilities.feature_collection

Source: flowmachine/features/utilities/feature_collection.py

Class definition for feature_collection, this is a group of joined features.

feature_collection

feature_collection(metrics, dropna=True) -> flowmachine.core.join.Join
Source: flowmachine/features/utilities/feature_collection.py

Joined set of features. Takes a set of features and creates one wide dataset about these features. Most often used to gather subscriber metrics into one dataframe, for instance to pass to a machine learning pipeline.

Parameters

  • metrics: list of Query type objects

    A list (or other iterable) of objects which derive from the flowmachine.Query base class.

  • dropna: bool

    Keeps rows in which a subscriber has some but not all of the features.

Returns

  • flowmachine.core.join.Join

    A Join object combining all the features

Examples

There are two alternative constructors to this class. The first is more general, it takes a list of query instances.

start, stop = '2016-01-01', '2016-01-03'
metrics = [RadiusOfGyration(start, stop),
               NocturnalCalls(start, stop),
               SubscriberDegree(start, stop)]


fc = feature_collection(metrics)
fc.head()
   subscriber            |rog_radiusofgyration_0|percentage_nocturnal_nocturnalcalls_1|degree_subscriberdegree_2
    ----------------|----------------------|-------------------------------------|-------------------
    038OVABN11Ak4W5P|162.003039769672      |28.571428571428573                   |2
    09NrjaNNvDanD8pk|191.563707606684      |58.333333333333336                   |2
    0ayZGYEQrqYlKw6g|253.993944670455      |27.272727272727273                   |2
    0DB8zw67E9mZAPK2|230.161989941767      |18.181818181818183                   |2
    0Gl95NRLjW2aw8pW|127.234294155594      |44.44444444444444                    |2
Sometimes a subscriber may only have a value associated to some of the features in the collection. The default behaviour of this class is only to return rows that have values for all the features. To override this behaviour we can do the following;
fc = feature_collection(metrics, dropna=False)
fc.head()
    subscriber      |rog_radiusofgyration_0|percentage_nocturnal_nocturnalcalls_1|degree_subscriberdegree_2
    ----------------|----------------------|-------------------------------------|-------------------
    038OVABN11Ak4W5P|162.003039769672      |28.571428571428573                   |2
    09NrjaNNvDanD8pk|Nan                   |58.333333333333336                   |2
    0ayZGYEQrqYlKw6g|253.993944670455      |Nan                                  |2
    0DB8zw67E9mZAPK2|230.161989941767      |18.181818181818183                   |2
    0Gl95NRLjW2aw8pW|127.234294155594      |44.44444444444444                    |2
An alternative, and easier way, to get the following is to do:
start, stop = '2016-01-01', '2016-01-03'
metrics = [RadiusOfGyration,
               NocturnalCalls,
               SubscriberDegree]


fc = feature_collection.feature_collection_from_list_of_classes(metrics, start, stop)
But this requires that you want the same arguments for each class (and are happy with the defaults).

Note

Each column has the name of the class appended to it to distinguish it from other potential inputs, and an integer. This is because the column names must be unique, and it is possible to use the same metric multiple times but with different parameters.

feature_collection_from_list_of_classes

feature_collection_from_list_of_classes(classes, *args, dropna=False, **kwargs) -> flowmachine.core.join.Join
Source: flowmachine/features/utilities/feature_collection.py

Create a feature collection from uninstantiated classes with common arguments.

Returns

  • flowmachine.core.join.Join

    A Join object combining all the features