finds.dev← search

// the find

facebook/prophet

★ 20,223 · Python · MIT · updated May 2026

Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.

Prophet is Facebook's time series forecasting library built on an additive model with piecewise linear/logistic trend, Fourier-series seasonality, and holiday effects. It's aimed at analysts and data scientists who need interpretable forecasts without deep statistics knowledge. Both Python and R are supported, with Stan doing the heavy lifting under the hood.

The decomposable model (trend + seasonality + holidays) means you can actually inspect what's driving each component — not a black box. Handling of missing data and outliers is genuinely good without manual intervention. The changepoint detection is automatic but also manually overridable, which is the right trade-off. Cross-validation tooling with time-series-aware cutoffs is built in and works correctly, which most DIY implementations get wrong.

Accuracy on non-business time series (anything without strong weekly/yearly seasonality) is often mediocre — ARIMA or even simple exponential smoothing will beat it. The Stan dependency makes installation painful on Windows and adds ~4GB RAM overhead at fit time. The project is effectively in maintenance mode: the 2023 blog post explicitly said Meta isn't actively developing it further, so you're adopting a frozen library. Uncertainty intervals assume the future looks like the past in ways that fail badly at structural breaks — the model will confidently forecast through COVID-style disruptions as if they're outliers.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →