The time series starts with the amount specified for the first application. This does not create objects of type ts.

one_box(x, t_end = 100, res = 0.01, ...)

# S3 method for numeric
one_box(x, t_end = 100, res = 0.01, ...)

# S3 method for character
one_box(x, parms, t_end = 100, res = 0.01, ...)

# S3 method for mkinfit
one_box(x, t_end = 100, res = 0.01, ...)

Arguments

x
When numeric, this is the half-life to be used for an exponential decline. If x is an mkinfit object, the decline is calculated from this object
t_end
End of the time series
res
Resolution of the time series
...
Further arguments passed to methods
parms
A named numeric vector containing the model parameters

Examples

# Only use a half-life pred_0 <- one_box(10) plot(pred_0)
# Use a fitted mkinfit model require(mkin) fit <- mkinfit("FOMC", FOCUS_2006_C, quiet = TRUE) pred_1 <- one_box(fit) plot(pred_1)
# Use a model with more than one observed variable m_2 <- mkinmod(parent = mkinsub("SFO", "m1"), m1 = mkinsub("SFO"))
#> Successfully compiled differential equation model from auto-generated C code.
fit_2 <- mkinfit(m_2, FOCUS_2006_D, quiet = TRUE) pred_2 <- one_box(fit_2) plot(pred_2)