Extensions and Speed Up

Vectorized keys

To speed up calculations, faas provides users with the concept of feature vectorization. Some parameters for some features can be written as a list of elements, where each element must be a valid parameter. In this case, instead of one feature, N features will be calculated, formed from the Cartesian product of the requested list of features. In the simplest case, this is a list of one element.

Vectorization allows you to calculate many features at the cost of calculating only one feature.

Simple example:

circle_feature = {
    'type': 'circle',
    'dataset': 'ru-pub-households/apartments',
    'field': ['apartments', 'price', 'rooms'],
    'limit': 300,
    'agg': ['sum', 'mean'],
    'fillna': 0,
}

As you can see, 2 fields are vectorized here: 'field': ['apartments', 'price', 'rooms'] and 'agg': ['sum', 'mean']. Thus, the result will be calculated 6 features as a Cartesian product of lists.

The following fields support vectorization: 'field', 'agg' for each type of feature, except Match_aggregate.

Since each feature must have a unique name, a template is used as the name of vectorized features. Example:

config = {
    ...,
    'features': {
        'apartments_{{FIELD}}_{{AGG}}': {
            'type': 'circle',
            'dataset': 'ru-pub-households/apartments',
            'field': ['price', 'year'],
            'limit': 300,
            'agg': ['mean', 'max'],
            'fillna': 0,
        }
    }
}

Based on the apartments_{{FIELD}}_{{AGG}} template, 4 features will be created:

• apartments_price_mean,
• apartments_year_mean,
• apartments_price_max,
• apartments_year_max.