Time series processing framework and utilities for deep learning
Project description
TimeWarPY - Time Series Pre and Post Processing Methods
Background and Objective
TimeWarPy is a library I created because I kept running into time-series related pre and post processing that is discussed a lot in ML literature but not standardized in a popular ML library. Most industry related forecasting methods are not well suited for real-time deep learning architectures. TimeWarPy is a stab at making these operations both fast and convenient for real-time applications through an easy to use set of core processing objections.
Installation
TimeWarPY can be installed directly with PyPi or directly from source here
pip install timewarpy
Motivation
Univariate Data
Time series data sets for deep learning generally need to be put in the visual format below. There will be a sequence in time (vector) for training and a prediction sequence in time (another vector) that is normally shorter.
This single example is then rolled in time to generate many examples of these training and predicting sequences as shown below.
There are a million variations to the rolling mechanism shown above including changing window sizes, increasing the number of time increments to roll, dealing with non-continuous time series, adding contextual information, dealing with time dependent meta data storage, and much more. TimeWarPY is intended to standardize these transformations to be ready for deep learning methods out of the box.
The example here: Preprocessing Univariate Data for Recurrent Neural Network Training gives a good indication how quickly TimeWarPY can get you training models with TensorFlow.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for timewarpy-0.0.18-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 5455f3fd2a11752c00759d37abbdb095a09a0d81f0747b907eef8d7e09be1e41 |
|
MD5 | d954fb7ee46a9062790791fbc17821fc |
|
BLAKE2b-256 | 9ab3a4c0f7728f3a7ec62df17a0a335e565a4147049fbccf470f3976039849e2 |