make_model_smoother {simITS}

Make a smoother that fits a model and then smooths residuals

Description

This helper function gives back a function that takes the resulting simulation data from a single iteration of the simulation, and fits 'fit_model' to it, smoothes the residuals, and puts the predictions from 'fit_model' back.

Usage

make_model_smoother(fit_model, covariates)

Arguments

fit_model

A function that takes data, fits a linear model, and returns the fit model. This function needs an option to include (or not) lagged covariates.

covariates

A dataframe with all covariates needed in the model fitting defined by fit_model.

Details

This can be used to build smoothers that smooth using models other than the model being used for extrapolation (e.g., a model without temperature).

Resulting functions have the following parameters: 'res' (the data), 't0' (start time), 'outcomename', 'post.only' flag (for smoothing only post data or not), and 'smooth_k', a tuning parameter for degree of smoothing.

Value

A smoother function that can be passed to the smoothing routines. This function is of the form listed above.

Examples

data( "newjersey")
modA = make_fit_season_model( ~ temperature )
modB = make_fit_season_model( ~ sin.m + cos.m )
newjersey = add_lagged_covariates( newjersey, "n.warrant", 
                                   covariates = c("sin.m", "cos.m", "temperature") )
smoother = make_model_smoother( fit_model = modA, covariates = newjersey )
class(smoother)

# Pass made function to process_outcome_model()
envelope = process_outcome_model( "n.warrant", newjersey, t0=-8, R = 1,
                                  summarize = TRUE, smooth=TRUE, 
                                  smoother = smoother, smooth_k = 11,
                                  fit.model = modB )