neural_prophet

Why we included this project

NeuralProphet suits teams that need to understand a forecast, not just receive one. It wraps PyTorch in a Prophet-style API, so a working model takes only a few lines of code, and you can decompose the fit into trend, seasonality, autoregressive, and regressor parts to see what each contributes. The real draw is the human-in-the-loop workflow: build a quick baseline, read the component plots and model coefficients, adjust, and repeat until the output matches your data. It handles higher-frequency series, trains across multiple related series at once, and returns quantile-based uncertainty ranges. Just note it is still in beta, so expect to iterate rather than treat it as a finished black box.