neural_prophet
NeuralProphet: A simple forecasting package
What is neural_prophet?
NeuralProphet: A simple forecasting package
Overview of neural_prophet
What it is
NeuralProphet is an open-source framework for interpretable time series forecasting. It is built on PyTorch and combines neural networks with traditional time-series algorithms, inspired by Facebook Prophet and AR-Net. The project is currently in beta phase.
Key features
- Model components include autoregression (linear or NN), piecewise linear trend with optional changepoint detection, Fourier-based seasonality, lagged regressors, future regressors, and events (holidays and custom events).
- Supports global, local, and 'glocal' (global + regularized local) modeling for multiple time series.
- Provides uncertainty estimation via quantile regression.
- Includes plotting functions for forecast components, model coefficients, and parameters.
- Offers time series crossvalidation utility, model checkpointing, and validation.
- Designed for iterative human-in-the-loop model building with interpretability and customization.
Use cases
NeuralProphet is best suited for time series data that is of higher-frequency (sub-daily) and longer duration (at least two full periods or years). It can be used for forecasting tasks where interpretability and iterative model refinement are important, such as demand forecasting, resource planning, or anomaly detection in time series.
Getting started
Install with pip:shellpip install neuralprophetFor Jupyter notebooks, install the 'live' version:shellpip install neuralprophet[live]To install the latest version from GitHub:shellgit clonecd neural_prophetpip install .Minimal usage example:pythonfrom neuralprophet import NeuralProphetm = NeuralProphet()metrics = m.fit(df)forecast = m.predict(df)
License and pricing
NeuralProphet is open-source and licensed under the MIT license. No pricing information is provided.
Based on the project's public README and repository metadata.