orbit: A Python repository from deepanshu-nickelfox

Disclaimer

This project

is stable and being incubated for long-term support. It may contain new experimental code, for which APIs are subject to change.
requires PyStan as a system dependency. PyStan is licensed under GPLv3, which is a free, copyleft license for software.

Orbit: A Python package for Bayesian forecasting models

Orbit is a Python package for Bayesian forecasting models developed under object-oriented design. It provides a familiar and intuitive initialize-fit-predict interface for working with time series tasks, while utilizing probabilistic modeling api under the hood.

The initial release supports concrete implementation for the following models:

Local Global Trend (LGT)
Damped Local Trend (DLT)

Both models, which are variants of exponential smoothing, support seasonality and exogenous (time-independent) features.

The initial release also supports the following sampling methods for model estimation:

Markov-Chain Monte Carlo (MCMC) as a full sampling method
Maximum a Posteriori (MAP) as a point estimate method
Variational Inference (VI) as a hybrid-sampling method on approximate distribution

Installation

Prerequisites

Install from PyPi:

pip install orbit-ml

Install from GitHub:

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

Quick Start

Load data

import pandas as pd
import numpy as np
from orbit.utils.dataset import load_iclaims

df = load_iclaims()
df[['claims', 'trend.unemploy', 'trend.filling', 'trend.job']] =\
    df[['claims', 'trend.unemploy', 'trend.filling', 'trend.job']].apply(np.log)

test_size=52
train_df=df[:-test_size]
test_df=df[-test_size:]

Damped-Local-Trend (DLT) Model with FULL Bayesian Prediction

from orbit.models.dlt import DLTFull
from orbit.diagnostics.plot import plot_predicted_data

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

# predicted df
predicted_df = dlt.predict(df=test_df)

# plot predictions
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
)

References

Documentation

Orbit API Documentation and Examples

Citation

To cite Orbit in publications, refer to the following whitepaper:

Orbit: Probabilistic Forecast with Exponential Smoothing

Bibtex:

@misc{ng2020orbit, title={Orbit: Probabilistic Forecast with Exponential Smoothing}, author={Edwin Ng, Zhishi Wang, Huigang Chen, Steve Yang, Slawek Smyl}, year={2020}, eprint={2004.08492}, archivePrefix={arXiv}, primaryClass={stat.CO}}

Papers

Hyndman, R., Koehler, A. B., Ord, J. K., and Snyder, R. D. Forecasting with exponential smoothing: the state space approach. Springer Science & Business Media, 2008.
Bingham, E., Chen, J. P., Jankowiak, M., Obermeyer, F., Pradhan, N., Karaletsos, T., Singh, R., Szerlip, P., Horsfall, P., and Goodman, N. D. Pyro: Deep universal probabilistic programming. The Journal of Machine Learning Research, 20(1):973–978, 2019.
Taylor, S. J. and Letham, B. Forecasting at scale. The American Statistician, 72(1):37–45, 2018.
Hoffman, M.D. and Gelman, A. The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. J. Mach. Learn. Res., 15(1), pp.1593-1623, 2014.

deepanshu-nickelfox/orbit