Compute S-map models for each time series in the block and
save out the coefficients. The coefficients represent the local linear
model and can be used to infer properties of the system dynamics.
compute_smap_coeffs(block, ccm_links, rescale = TRUE, rolling_forecast = FALSE, id_var = NULL)
| block | a data.frame containing time series for the community. Each column is a time series of abundances. |
|---|---|
| ccm_links | A data.frame containing the significant causal links. Each row is a causal link. The columns are:
|
| rescale | A logical, indicating whether to rescale each time series |
| rolling_forecast | A logical, indicating whether to make individual rolling forecasts for the second half of the time series. |
| id_var | the name of the index variable of |
A list with the matrix smap-coefficients for each predictor variable identified in CCM (these are the affected variables). The names in the list and the column names of the matrices use the variable names in the block.
Suppose that there are causal links as follows:
x --> y
x --> z
where --> indicates "cross-maps to". Then the interpretation from CCM is
that x is affected by causes y and z. Thus, the predictive model for
x should include y and z as predictors.
The S-map model is then setup as
x_t+1 = F(x_t, y_t, z_t, x_t-1, x_t-2, ...)
where the number of predictors is equal to the best embedding dimension for
x.