Applies a conservative switching gate to the first row of res$table.
The best candidate must pass residual QS and Ljung-Box thresholds, must not
be too far from the incumbent relative to the candidate set, and must retain
sufficient seasonal-component correlation when that comparison is available.
AICc, revision metrics and engine preference affect the ranking in
auto_seasonal_analysis(), but are not directly re-applied by this switching
helper.
Usage
sa_should_switch(
res,
thresholds = list(min_qs_p = 0.1, max_dist_sa_mult = 1.25, min_corr_seas = 0.9,
min_lb_p = 0.05)
)Arguments
- res
Result of
auto_seasonal_analysis().- thresholds
Named list with decision thresholds:
min_qs_p: minimum acceptable QS p-value on SA (overall) for the best model
max_dist_sa_mult: allow SA L1 distance up to this multiple of the cross-candidate median
min_corr_seas: minimum correlation of seasonal components (vs. incumbent)
min_lb_p: minimum acceptable Ljung-Box p-value on residuals
Examples
# \donttest{
if (requireNamespace("seasonal", quietly = TRUE)) {
res <- auto_seasonal_analysis(AirPassengers, max_specs = 3)
sa_should_switch(res)
}
#> Model used in SEATS is different: (1 1 2)(1 0 0)
#> Model used in SEATS is different: (1 1 2)(1 0 0)
#> Model used in SEATS is different: (1 1 2)(1 0 0)
#> [1] "CHANGE_TO_NEW_MODEL"
# }