https://github.com/octoenergy/timeserio

Better `keras` models for time series and beyond
https://github.com/octoenergy/timeserio

data data-science

Last synced: about 1 year ago
JSON representation

Better `keras` models for time series and beyond

• Host: GitHub
• URL: https://github.com/octoenergy/timeserio
• Owner: octoenergy
• License: mit
• Created: 2019-07-11T19:18:04.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2024-01-11T19:43:45.000Z (over 2 years ago)
• Last Synced: 2025-06-10T15:07:21.328Z (about 1 year ago)
• Topics: data, data-science
• Language: Python
• Size: 9.44 MB
• Stars: 61
• Watchers: 97
• Forks: 16
• Open Issues: 13
• Metadata Files:
  • Readme: README.md
  • Contributing: CONTRIBUTING.md
  • License: LICENSE.txt

Awesome Lists containing this project

README

          
[![CircleCI](https://circleci.com/gh/octoenergy/timeserio/tree/master.svg?style=svg)](https://circleci.com/gh/octoenergy/timeserio/tree/master) 

[![codecov](https://codecov.io/gh/octoenergy/timeserio/branch/master/graph/badge.svg)](https://codecov.io/gh/octoenergy/timeserio)

[![PyPI version](https://badge.fury.io/py/timeserio.svg)](https://badge.fury.io/py/timeserio)

# timeserio

`timeserio` is the missing link between `pandas`, `scikit-learn` and `keras`. It simplifies building end-to-end deep learning models - from a DataFrame through feature pipelines to multi-stage models with shared layers. While initially developed for tackling time series problems, it has since been used as a versatile tool for rapid ML model development and deployment.

Loosing track of big networks with multiple inputs and outputs? Forgetting to freeze the right layers?

Struggling to re-generate the input features? `timeserio` can help!

![complex_network](https://raw.githubusercontent.com/octoenergy/timeserio/master/docs/source/_static/multinetwork_complex.svg?sanitize=true)

## Documentation and Tutorials

Please see the [official documentation](http://tech.octopus.energy/timeserio/) on how to get started.

## Features

* Enable encapsulated, maintainable and reusable deep learning models

* Feed data from `pandas` through `scikit-learn` feature pipelines to multiple neural network inputs

* Manage complex architectures, layer sharing, partial freezing and re-training

* Provide collection of extensible building blocks with emphasis on time series problems

## Installation

`pip install timeserio`, or install from source - `pip install -e .`

See [Getting Started](http://tech.octopus.energy/timeserio/overview/getting_started.html#installation)

## Development

We welcome contributions and enhancements to any part of the code base, documentation, or tool chain.

See [CONTRIBUTING.md](https://github.com/octoenergy/timeserio/blob/master/CONTRIBUTING.md) for details on setting up the development environment, running tests,

etc.