Project: statsforecast

Time series forecasting suite using statistical models

Project Details

Latest version
1.7.0
Home Page
https://github.com/Nixtla/statsforecast/
PyPI Page
https://pypi.org/project/statsforecast/

Project Popularity

PageRank
0.00760700969590256
Number of downloads
469548

Nixtla Β  Tweet Β Slack

All Contributors

Statistical ⚑️ Forecast

Lightning fast forecasting with statistical and econometric models

CI Python PyPi conda-nixtla License docs Downloads

StatsForecast offers a collection of widely used univariate time series forecasting models, including automatic ARIMA, ETS, CES, and Theta modeling optimized for high performance using numba. It also includes a large battery of benchmarking models.

Installation

You can install StatsForecast with:

pip install statsforecast

or

conda install -c conda-forge statsforecast

Vist our Installation Guide for further instructions.

Quick Start

Minimal Example

from statsforecast import StatsForecast
from statsforecast.models import AutoARIMA
from statsforecast.utils import AirPassengersDF

df = AirPassengersDF
sf = StatsForecast(
    models = [AutoARIMA(season_length = 12)],
    freq = 'M'
)

sf.fit(df)
sf.predict(h=12, level=[95])

Get Started with this quick guide.

Follow this end-to-end walkthrough for best practices.

Why?

Current Python alternatives for statistical models are slow, inaccurate and don't scale well. So we created a library that can be used to forecast in production environments or as benchmarks. StatsForecast includes an extensive battery of models that can efficiently fit millions of time series.

Features

  • Fastest and most accurate implementations of AutoARIMA, AutoETS, AutoCES, MSTL and Theta in Python.
  • Out-of-the-box compatibility with Spark, Dask, and Ray.
  • Probabilistic Forecasting and Confidence Intervals.
  • Support for exogenous Variables and static covariates.
  • Anomaly Detection.
  • Familiar sklearn syntax: .fit and .predict.

Highlights

  • Inclusion of exogenous variables and prediction intervals for ARIMA.
  • 20x faster than pmdarima.
  • 1.5x faster than R.
  • 500x faster than Prophet.
  • 4x faster than statsmodels.
  • Compiled to high performance machine code through numba.
  • 1,000,000 series in 30 min with ray.
  • Replace FB-Prophet in two lines of code and gain speed and accuracy. Check the experiments here.
  • Fit 10 benchmark models on 1,000,000 series in under 5 min.

Missing something? Please open an issue or write us in Slack

Examples and Guides

πŸ“š End to End Walkthrough: Model training, evaluation and selection for multiple time series

πŸ”Ž Anomaly Detection: detect anomalies for time series using in-sample prediction intervals.

πŸ‘©β€πŸ”¬ Cross Validation: robust model’s performance evaluation.

❄️ Multiple Seasonalities: how to forecast data with multiple seasonalities using an MSTL.

πŸ”Œ Predict Demand Peaks: electricity load forecasting for detecting daily peaks and reducing electric bills.

πŸ“ˆ Intermittent Demand: forecast series with very few non-zero observations.

🌑️ Exogenous Regressors: like weather or prices

Models

Automatic Forecasting

Automatic forecasting tools search for the best parameters and select the best possible model for a group of time series. These tools are useful for large collections of univariate time series.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
AutoARIMA βœ… βœ… βœ… βœ… βœ…
AutoETS βœ… βœ… βœ… βœ…
AutoCES βœ… βœ… βœ… βœ…
AutoTheta βœ… βœ… βœ… βœ…

ARIMA Family

These models exploit the existing autocorrelations in the time series.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
ARIMA βœ… βœ… βœ… βœ… βœ…
AutoRegressive βœ… βœ… βœ… βœ… βœ…

Theta Family

Fit two theta lines to a deseasonalized time series, using different techniques to obtain and combine the two theta lines to produce the final forecasts.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
Theta βœ… βœ… βœ… βœ…
OptimizedTheta βœ… βœ… βœ… βœ…
DynamicTheta βœ… βœ… βœ… βœ…
DynamicOptimizedTheta βœ… βœ… βœ… βœ…

Multiple Seasonalities

Suited for signals with more than one clear seasonality. Useful for low-frequency data like electricity and logs.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
MSTL βœ… βœ… βœ… βœ… If trend forecaster supports

GARCH and ARCH Models

Suited for modeling time series that exhibit non-constant volatility over time. The ARCH model is a particular case of GARCH.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
GARCH βœ… βœ… βœ… βœ…
ARCH βœ… βœ… βœ… βœ…

Baseline Models

Classical models for establishing baseline.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
HistoricAverage βœ… βœ… βœ… βœ…
Naive βœ… βœ… βœ… βœ…
RandomWalkWithDrift βœ… βœ… βœ… βœ…
SeasonalNaive βœ… βœ… βœ… βœ…
WindowAverage βœ…
SeasonalWindowAverage βœ…

Exponential Smoothing

Uses a weighted average of all past observations where the weights decrease exponentially into the past. Suitable for data with clear trend and/or seasonality. Use the SimpleExponential family for data with no clear trend or seasonality.

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
SimpleExponentialSmoothing βœ…
SimpleExponentialSmoothingOptimized βœ…
SeasonalExponentialSmoothing βœ…
SeasonalExponentialSmoothingOptimized βœ…
Holt βœ… βœ… βœ… βœ…
HoltWinters βœ… βœ… βœ… βœ…

Sparse or Intermittent

Suited for series with very few non-zero observations

Model Point Forecast Probabilistic Forecast Insample fitted values Probabilistic fitted values Exogenous features
ADIDA βœ…
CrostonClassic βœ…
CrostonOptimized βœ…
CrostonSBA βœ…
IMAPA βœ…
TSB βœ…

πŸ”¨ How to contribute

See CONTRIBUTING.md.

Citing

@misc{garza2022statsforecast,
    author={Federico Garza, Max Mergenthaler Canseco, Cristian ChallΓΊ, Kin G. Olivares},
    title = {{StatsForecast}: Lightning fast forecasting with statistical and econometric models},
    year={2022},
    howpublished={{PyCon} Salt Lake City, Utah, US 2022},
    url={https://github.com/Nixtla/statsforecast}
}

Contributors ✨

Thanks goes to these wonderful people (emoji key):

fede
fede

πŸ’» 🚧
JosΓ© Morales
JosΓ© Morales

πŸ’» 🚧
Sugato Ray
Sugato Ray

πŸ’»
Jeff Tackes
Jeff Tackes

πŸ›
darinkist
darinkist

πŸ€”
Alec Helyar
Alec Helyar

πŸ’¬
Dave Hirschfeld
Dave Hirschfeld

πŸ’¬
mergenthaler
mergenthaler

πŸ’»
Kin
Kin

πŸ’»
Yasslight90
Yasslight90

πŸ€”
asinig
asinig

πŸ€”
Philip Gillißen
Philip Gillißen

πŸ’»
Sebastian Hagn
Sebastian Hagn

πŸ› πŸ“–
Han Wang
Han Wang

πŸ’»
Ben Jeffrey
Ben Jeffrey

πŸ›
Beliavsky
Beliavsky

πŸ“–
Mariana Menchero GarcΓ­a
Mariana Menchero GarcΓ­a

πŸ’»
Nikhil Gupta
Nikhil Gupta

πŸ›
JD
JD

πŸ›
josh attenberg
josh attenberg

πŸ’»
JeroenPeterBos
JeroenPeterBos

πŸ’»
Jeroen Van Der Donckt
Jeroen Van Der Donckt

πŸ’»
Roymprog
Roymprog

πŸ“–
Nelson CΓ‘rdenas BolaΓ±o
Nelson CΓ‘rdenas BolaΓ±o

πŸ“–
Kyle Schmaus
Kyle Schmaus

πŸ’»
Akmal Soliev
Akmal Soliev

πŸ’»
Nick To
Nick To

πŸ’»
Kevin Kho
Kevin Kho

πŸ’»
Yiben Huang
Yiben Huang

πŸ“–
Andrew Gross
Andrew Gross

πŸ“–
taniishkaaa
taniishkaaa

πŸ“–
Manuel Calzolari
Manuel Calzolari

πŸ’»

This project follows the all-contributors specification. Contributions of any kind welcome!