Introduction to Orbit

Orbit is a cutting-edge Python package designed for Bayesian time series forecasting and inference. It provides a user-friendly interface that simplifies the process of initializing, fitting, and predicting time series data, all while utilizing advanced probabilistic programming techniques under the hood.

With its robust architecture, Orbit supports various models and sampling methods, making it an essential tool for data scientists and developers working with time series data.

Main Features of Orbit

• Multiple Model Support: Implementations for Exponential Smoothing (ETS), Local Global Trend (LGT), Damped Local Trend (DLT), and Kernel Time-based Regression (KTR).
• Flexible Sampling Methods: Choose from Markov-Chain Monte Carlo (MCMC), Maximum a Posteriori (MAP), and Variational Inference (VI) for model estimation.
• Intuitive API: A straightforward initialize-fit-predict interface that streamlines the forecasting process.
• Comprehensive Documentation: Extensive tutorials and examples to help users get started quickly.

Technical Architecture and Implementation

Orbit is built on a solid foundation that leverages probabilistic programming languages to provide accurate forecasting capabilities. The library is designed to be modular, allowing users to easily extend its functionality or integrate it with other tools.

Key components of Orbit include:

• Model Templates: Define the structure of various forecasting models.
• Estimators: Implement the logic for fitting models to data.
• Forecasters: Handle the prediction process and output results.

Setup and Installation Process

Installing Orbit is straightforward. You can choose to install the stable release or the development version based on your needs.

Installing Stable Release

To install Orbit from PyPI, use the following command:

$ pip install orbit-ml

Alternatively, you can install it from source:

$ git clone https://github.com/uber/orbit.git
$ cd orbit
$ pip install -r requirements.txt
$ pip install .

For Anaconda users, install from the conda-forge channel:

$ conda install -c conda-forge orbit-ml

Installing from Dev Branch

If you want to use the latest features from the development branch, run:

$ pip install git+https://github.com/uber/orbit.git@dev

Usage Examples and API Overview

Orbit provides a variety of models for time series forecasting. Below is a quick start example using the Damped Local Trend (DLT) model:

from orbit.utils.dataset import load_iclaims
from orbit.models import DLT
from orbit.diagnostics.plot import plot_predicted_data

# Load log-transformed data
df = load_iclaims()
# Train-test split
test_size = 52
train_df = df[:-test_size]
test_df = df[-test_size:]

dlt = DLT(
  response_col='claims', date_col='week',
  regressor_col=['trend.unemploy', 'trend.filling', 'trend.job'],
  seasonality=52,
)
dlt.fit(df=train_df)

# Outcomes data frame
predicted_df = dlt.predict(df=test_df)

plot_predicted_data(
  training_actual_df=train_df, predicted_df=predicted_df,
  date_col=dlt.date_col, actual_col=dlt.response_col,
  test_actual_df=test_df
)

This example demonstrates how to load data, fit a model, and visualize the predictions.

Community and Contribution Aspects

Orbit is an open-source project that welcomes contributions from the community. If you’re interested in contributing, please review the Code of Conduct and the Contributing Guidelines.

Before contributing, check out the outstanding issues to see how you can help.

License and Legal Considerations

Orbit is licensed under the Apache License, Version 2.0. This allows you to use, modify, and distribute the software under certain conditions. Make sure to review the license for compliance.

Project Roadmap and Future Plans

Orbit is continuously evolving, with plans for new features and improvements. The team is actively working on enhancing the documentation, adding more models, and improving the overall user experience. Stay tuned for updates!

Conclusion

Orbit is a powerful tool for anyone looking to implement Bayesian forecasting in their projects. With its intuitive interface and robust features, it simplifies the complexities of time series analysis. Whether you’re a seasoned data scientist or just starting, Orbit provides the tools you need to succeed.

For more information, visit the official documentation or check out the GitHub repository.

FAQ Section