add user guide info on forecasting ahead of observed data about redm

[atredennick]

@ha0ye Can rEDM::simplex() make forecasts of unobserved future states when modeling a univariate (i.e., no input data) time series? I've tried tacking on NA to the state vector, but this doesn't seem to work. Thanks!

[ha0ye]

Hi @atredennick, are you trying to do a 1-step ahead forecast or a whole series of forecasts to simulate a future trajectory?

If it's the former, you might need some tweaking of the pred params, so that there are enough lags to form the state vector to predict from:

ts <- c(rnorm(54), NA)
lib <- c(1, 50)   # use points 1-50 to construct model
pred <- c(51, 55) # use points 51-54 to construct state vector, 
                  # predict point 55

out <- simplex(ts, lib = lib, pred = pred, 
               E = 4, stats_only = FALSE)
out$model_output

[atredennick]

Thank you @ha0ye! That works great. Yes, I am just trying to do 1-step ahead forecasts at the moment. Ideally, I'd actually just like to just simulate time-series from the fitted model from, say, different initial conditions. Looks like the best way to do that with rEDM is to iteratively forecast. But maybe I'm wrong.

Thanks again.

[ha0ye]

Great! Yeah, the code is built to maintain similar operations as legacy code, so it doesn't function in quite the same way as other kinds of models in R, where the model object can be used directly to iteratively make forecasts. You'll probably want to write some kind of wrapper function to make that easier.