Prophet is a time series forecasting library based on additive models and nonlinear trends. When using Prophet to make predictions, some parameters need to be set to control the behavior of the model. The following are some commonly used parameters in Prophet:
-
growth : Determine the growth trend of the time series, you can choose linear or nonlinear. The default is linear trend.
-
seasonality_mode : Controls how the seasonality component is modeled, additive or multiplicative. The default is additive.
-
changepoint_prior_scale : Controls the sensitivity of changepoints. Higher values indicate that the model adapts more easily to changes, but can also lead to overfitting.
-
seasonality_prior_scale : Controls the sensitivity of the seasonality component. Higher values indicate that the model fits seasonality more easily, but can also lead to overfitting.
-
holidays_prior_scale : Controls the sensitivity of the holidays effect. Higher values indicate that the model adapts more easily to holiday effects, but can also lead to overfitting.
-
interval_width : The width of the confidence interval, indicating the accuracy of the prediction. The default is 0.8, which is the 80% confidence interval.
-
n_changepoints : Used to determine the number of changepoints. The default is 25.
-
seasonality : Controls the cycle of seasonality. Can be an integer representing the number of days in the period, or a string representing "years", "quarters", "months", "weeks", "days", "hours", "minutes", or "seconds". Default is yearly.
-
holidays : A data frame of holiday dates, needs to contain two columns "holiday" and "ds". Default is empty.
-
weekly_seasonality : Controls how weekly seasonality is modeled, can be 'auto', 'True', 'False' or an integer. The default is "auto", which means to automatically detect weekly seasonality in the data.
-
daily_seasonality : Controls how daily seasonality is modeled, can be 'auto', 'True', 'False' or an integer. The default is "auto", meaning that daily seasonality is automatically detected in the data.
-
yearly_seasonality : Controls how yearly seasonality is modeled, can be 'auto', 'True', 'False' or an integer. The default is "auto", which means to automatically detect annual seasonality in the data.
-
seasonality_prior_mode : Controls how the seasonality prior is modeled. You can choose from "auto", "additive", or "multiplicative". The default is "auto".
-
uncertainty_samples : Controls the number of samples for uncertainty estimation. The default is 1000.
The above parameters are some commonly used parameters when using Prophet for time series forecasting. These parameters can be adjusted appropriately according to actual needs to obtain better forecasting results.