NeuralProphet

Based on Neural Networks, inspired by Facebook Prophet and AR-Net, built on Pytorch.

Links

• Read the paper

• GitHub repository

Why NeuralProphet?

NeuralProphet changes the way time series modelling and forecasting is done:

• Support for auto-regression and covariates.

• Automatic selection of training related hyperparameters.

• Fourier term seasonality at different periods such as yearly, daily, weekly, hourly.

• Piecewise linear trend with optional automatic changepoint detection.

• Plotting for forecast components, model coefficients and final predictions.

• Support for global modeling.

• Lagged and future regressors.

• Sparsity of coefficients through regularization.

• User-friendly and powerful Python package:

>>> from neuralprophet import NeuralProphet
>>> m = NeuralProphet()
>>> metrics = m.fit(your_df, freq='D')
>>> forecast = m.predict(your_df)
>>> m.plot(forecast)

Installing

NeuralProphet can be installed with pip:

$ pip install neuralprophet

If you plan to use the package in a Jupyter notebook, we recommend to install the ‘live’ version:

$ pip install neuralprophet[live]

Alternatively, you can get the most up to date version by cloning directly from GitHub:

$ git clone https://github.com/ourownstory/neural_prophet.git
$ cd neural_prophet
$ pip install .

Get started with Tutorials

Feature Tutorials

• Autoregression
• Running benchmarking experiments
• Prediction Collection
• Test and CrossValidate
• Using Lagged Covariates
• Modeling Holidays and Special Events
• Multiplicative Seasonality
• Sparse Autoregression
• Sub-daily data
• Fitting a changing trend

Application Tutorials

• Building load forecasting: Hospital in SF
• Renewable Energy: Forecasting hourly solar irradiance