timeseriesfeatures 0.1.2

A library for processing timeseries features over a dataframe.

timeseries-features

A library for processing timeseries features over a dataframe of timeseries.

Dependencies :globe_with_meridians:

Python 3.11.6:

• pandas
• pyarrow
• statsmodels
• numpy
• sktime
• pytorch-forecasting
• mlflow
• litmodels
• dill
• cloudpickle
• torch
• pmdarima

Raison D'être :thought_balloon:

timeseries-features aims to process features relevant to predicting future values.

Architecture :triangular_ruler:

timeseries-features is a functional library, meaning that each phase of feature extraction gets put through a different function until the final output. It contains a feature extractor for auto-features, such as:

1. Autocorrelation
2. Partial Autocorrelation
3. NBEATS Predictions
4. ARIMA Predictions

To determine the auto-features it can use the following transforms:

1. No-Op
2. Velocity
3. Log
4. Acceleration
5. Jerk
6. Snap
7. SMA 5
8. Crackle

And a feature processor for the following features:

1. Lags
2. Rolling Count
3. Rolling Sum
4. Rolling Mean
5. Rolling Median
6. Rolling Variance
7. Rolling Standard Deviation
8. Rolling Minimum
9. Rolling Maximum
10. Rolling Skew
11. Rolling Kurtosis
12. Rolling Standard Error of the Mean
13. Rolling Rank
14. NBEATS Forecast
15. ARIMA Forecast

Installation :inbox_tray:

This is a python package hosted on pypi, so to install simply run the following command:

pip install timeseriesfeatures

or install using this local repository:

python setup.py install --old-and-unmanageable

Usage example :eyes:

The use of timeseriesfeatures is entirely through code due to it being a library. It attempts to hide most of its complexity from the user, so it only has a few functions of relevance in its outward API.

Generating Features

To generate features:

import datetime

import pandas as pd

from timeseriesfeatures.process import compute
from timeseriesfeatures.process import process
from timeseriesfeatures.feature import Feature, FEATURE_TYPE_LAG, FEATURE_TYPE_ROLLING, VALUE_TYPE_NONE, VALUE_TYPE_DAYS

df = ... # Your timeseries dataframe
features = compute(df, max_lag=30)
features.extend([
    Feature(feature_type=FEATURE_TYPE_LAG, value1=1),
    Feature(feature_type=FEATURE_TYPE_LAG, value1=2),
    Feature(feature_type=FEATURE_TYPE_LAG, value1=4),
    Feature(feature_type=FEATURE_TYPE_LAG, value1=8),
    Feature(feature_type=FEATURE_TYPE_ROLLING, value1=VALUE_TYPE_NONE, value2=None),
    Feature(feature_type=FEATURE_TYPE_ROLLING, value1=VALUE_TYPE_DAYS, value2=30),
])
df = process(df, features=features)

This will produce a dataframe that contains the new timeseries related features.

License :memo:

The project is available under the MIT License.