TSFresh Primitives

Installation

Install with pip:

python -m pip install "featuretools[tsfresh]"

Calculating Features

In tsfresh, this is how you can calculate a feature.

from tsfresh.feature_extraction.feature_calculators import agg_autocorrelation

data = list(range(10))
param = [{'f_agg': 'mean', 'maxlag': 5}]
agg_autocorrelation(data, param=param)

[('f_agg_"mean"__maxlag_5', 0.1717171717171717)]

With tsfresh primtives in featuretools, this is how you can calculate the same feature.

from featuretools.tsfresh import AggAutocorrelation

data = list(range(10))
AggAutocorrelation(f_agg='mean', maxlag=5)(data)

0.1717171717171717

Combining Primitives

In featuretools, this is how to combine tsfresh primitives with built-in or other installed primitives.

import featuretools as ft
from featuretools.tsfresh import AggAutocorrelation, Mean

entityset = ft.demo.load_mock_customer(return_entityset=True)
agg_primitives = [Mean, AggAutocorrelation(f_agg='mean', maxlag=5)]
feature_matrix, features = ft.dfs(entityset=entityset, target_dataframe_name='sessions', agg_primitives=agg_primitives)

feature_matrix[[
    'MEAN(transactions.amount)',
    'AGG_AUTOCORRELATION(transactions.amount, f_agg=mean, maxlag=5)',
]].head()

            MEAN(transactions.amount)  AGG_AUTOCORRELATION(transactions.amount, f_agg=mean, maxlag=5)
session_id
1                           76.813125                                           0.044268
2                           74.696000                                          -0.053110
3                           88.600000                                           0.007520
4                           64.557200                                          -0.034542
5                           70.638182                                          -0.100571

Notice that tsfresh primtives are applied across relationships in an entityset generating many features that are otherwise not possible.

feature_matrix[['customers.AGG_AUTOCORRELATION(transactions.amount, f_agg=mean, maxlag=5)']].head()

            customers.AGG_AUTOCORRELATION(transactions.amount, f_agg=mean, maxlag=5)
session_id
1                                                    0.011102
2                                                   -0.001686
3                                                   -0.010679
4                                                    0.011204
5                                                   -0.010679