Prophet

Prophet decomposes time series into trend, seasonality, and holiday effects. It is designed to handle missing data, outliers, and trend changes robustly, with interpretable components.

When to use:

• Business time series with strong seasonal effects and holiday impacts
• Forecasting when the series has trend shifts (changepoints)
• When interpretable trend and seasonality decomposition is needed
• Analysts who want to incorporate domain knowledge (custom holidays, capacity bounds)

Input:

• Trained model checkpoint — exported Prophet model
• Preprocessing config — date/feature normalization settings
• Steps — number of future periods to forecast

Output: Forecasted values with confidence intervals, trend, and seasonality components

Model Settings (set during training, used at inference)

Seasonality Mode (default: additive) additive — seasonal fluctuations are constant; multiplicative — fluctuations scale with trend level.

Changepoint Prior Scale (default: 0.05) Flexibility of the trend changepoints. Higher values allow more abrupt trend changes.

Seasonality Prior Scale (default: 10.0) Strength of the seasonality effect.

Holidays Prior Scale (default: 10.0) Magnitude of holiday effects.

Growth (default: linear) Trend model. linear for unbounded growth; logistic for saturating growth with a defined capacity.

Inference Settings

No dedicated inference-time settings. Prophet generates future dates automatically based on the specified number of periods.