NeoProphet

Prophet forecasting that is easier to use, clearer to read and faster at scale.

NeoProphet is a drop-in fork of Facebook Prophet with a one-call API, self-explaining charts, built-in tuning and an experimental Apple Silicon GPU backend for fitting many series at once.

Install

pip install neoprophet            # or: uv add neoprophet
pip install "neoprophet[mlx]"     # adds the MLX backend on Apple Silicon

NeoProphet installs the prophet import package, so it replaces upstream Prophet: uninstall prophet first and do not install both in one environment. Wheels include the compiled Stan model for macOS (Intel and Apple Silicon), Linux (x86-64 and ARM64) and Windows.

Quick start

import pandas as pd
from prophet import forecast

df = pd.read_csv("https://raw.githubusercontent.com/CodeWithBehnam/NeoProphet/main/examples/example_wp_log_peyton_manning.csv")
result = forecast(df, horizon="12 months", country_holidays="US")
result.forecast            # ds, yhat, yhat_lower, yhat_upper for the future
result.plot()              # the chart above
result.plot_components()   # trend, holidays, weekly and yearly patterns

• forecast(..., tune=True) chooses the prior scales and seasonality mode by parallel cross-validation.
• forecast(..., tune="optuna") searches ranges for the two prior scales, and both seasonality modes, with Optuna instead (pip install "neoprophet[optuna]"). The Optuna study is result.tuning.study. For your own ranges, call prophet.tune(df, param_grid=..., search="optuna", n_trials=40) with FloatRange and IntRange.
• prophet.mlx_backend.fit_many(dfs) fits many series in one batch on an Apple Silicon GPU (experimental; linear and flat growth, MAP only).
• The classic API is unchanged: from prophet import Prophet works as before.

Links

• Source, benchmarks and notebooks: https://github.com/CodeWithBehnam/NeoProphet
• Issues: https://github.com/CodeWithBehnam/NeoProphet/issues
• Upstream Prophet documentation: https://facebook.github.io/prophet/
• Paper: Taylor and Letham (2018), Forecasting at Scale, The American Statistician 72(1):37-45.

Released under the MIT licence. Prophet was created by Sean J. Taylor and Ben Letham at Facebook.