feature_encoders.compose package

Module contents

class feature_encoders.compose._compose.FeatureComposer(model_structure: feature_encoders.compose._compose.ModelStructure)[source]

Generate linear features and pairwise interactions.

Parameters: model_structure (ModelStructure) – The structure of a linear regression model.

property component_matrix

Dataframe indicating which columns of the feature matrix correspond to which components.

Returns: feature_cols – in that component.
Return type: A binary indicator dataframe. Entry is 1 if that column is used

fit(X, y=None)[source]

transform(X)[source]

class feature_encoders.compose._compose.ModelStructure(structure: Optional[Dict] = None, feature_map: Optional[Dict] = None)[source]

Capture the structure of a linear regression model.

The class validates and stores the details of a linear regression model: features, main effects and interactions.

Parameters

structure (Dict, optional) –

A dictionary that includes information about the model. Example:

{'add_features':
    {'time':
        { 'ds': None,
        'remainder': 'passthrough',
        'replace': False,
        'subset': ['month', 'hourofweek']
        }
    },
'main_effects':
    {'month':
        { 'feature': 'month',
        'max_n_categories': None,
        'encode_as': 'onehot',
        'interaction_only': False
        },
    'tow':
        { 'feature': 'hourofweek',
        'max_n_categories': 60,
        'encode_as': 'onehot',
        'interaction_only': False
        },
    'lin_temperature':
        { 'feature': 'temperature',
        'include_bias': False,
        'interaction_only': False
        }
    },
}

Defaults to None.

feature_map (Dict, optional) –
A mapping between a feature generator name and the classes for its validation and creation. Example:
```
{'datetime':
    'validate': 'validate.DatetimeSchema'
    'generate': 'generate.DatetimeFeatures'
}
```
Defaults to None.

add_interaction(*, lenc_name: str, renc_name: str, lenc_type: Union[str, object], renc_type: Union[str, object], **kwargs)[source]

Add a pairwise interaction.

Parameters

lenc_name (str) – A name for the first part of the interaction pair.
renc_name (str) – A name for the second part of the interaction pair.
lenc_type (str or encoder object) – The type of the feature encoder to apply on the first part of the interaction pair.
renc_type (str or encoder object) – The type of the feature encoder to apply on the second part of the interaction pair.
**kwargs – Keyword arguments to be passed during the feature encoders’ initialization.

Raises

ValueError – If an interaction with the same name (lenc_name, renc_name) has already been added.

Returns

The updated ModelStructure instance.

Return type

ModelStructure

Example:

model = ModelStructure().add_interaction(
    lenc_name="is_Monday",
    renc_name="daily_seasonality",
    lenc_type="categorical",
    renc_type="linear",
    **{
        is_Monday: {"feature": "is_Monday", "encode_as": "onehot"},
        daily_seasonality: {"feature": "daily", "as_filter": True},
    },
)

add_main_effect(*, name: str, enc_type: Union[str, sklearn.base.BaseEstimator], **kwargs)[source]

Add a main effect.

Parameters

name (str) – A name for the main effect.
enc_type (str or encoder object) – The type of the feature encoder to apply on the main effect.
**kwargs – Keyword arguments to be passed during the feature encoder initialization. Ignored if enc_type is not a string.

Raises

ValueError – If an encoder with the same name has already been added.

Returns

The updated ModelStructure instance.

Return type

ModelStructure

add_new_feature(*, name: str, fgen_type: Union[str, sklearn.base.BaseEstimator], **kwargs)[source]

Add a feature generator.

Feature generators are applied on the input dataframe with the same order that they were added.

Parameters

name (str) – A name for the feature generator.
fgen_type (str or sklearn-compatible transformer) – The feature generator to add. If it is a string, the corresponding class will be loaded based on the relevant entry in the feature_map dictionary.
**kwargs – Keyword arguments to be passed during the feature generator initialization. Ignored if fgen is not a string.

Raises

ValueError – If a feature generator with the same name has already been added.

Returns

The updated ModelStructure instance.

Return type

ModelStructure

property components

classmethod from_config(config: Dict, feature_map: Optional[Dict] = None)[source]

Create a ModelStructure instance from a configuration file.

Parameters

config (Dict) – A dictionary that includes information about the model.
feature_map (Dict, optional) – A mapping between a feature generator name and the classes for its validation and creation. Defaults to None.

Returns

A populated ModelStructure instance.

Return type

ModelStructure