Performance: Performance metrics
In oswaldosantos/INLAOutputs: Process selected outputs from the 'INLA' package

Description Usage Arguments Details Value Examples

Calculates performance metrics for regression and classification models..

1 2	Performance(..., samples = NULL, regression = TRUE, threshold = NULL, rnd = 3)

`...`	objects of `class` `inla` with computed fitted marginals.
`samples`	`NULL` or `list` with resamples. If `NULL`, performance metrics are calculated using the entaire dataset. If `list`, each element should be a `vector` with indexes to train the model. Indexes out of the vector (out-of-bag samples) are used to calculate performance metrics.
`regression`	logical. If `TRUE` (default), the RMSE and R2 are calculated. If `FALSE`, the following mwtrics are calculated: accuracy, sensitivity, specificity, Positive Predictive Value (PPV) and Negative Predictive Value (NPV).
`threshold`	When `regression = FALSE`, the threshold to assign class probabilities to classes is the value that maximizes `max(sensitivities + specificities)` (see details). Class probabilities below the threshold are assigned to `min(y)`, where `y` is the predicted variable (class probabilities above the threshold are assigned to the other class). If `threshold` is provided, it overrides the calculated threshold; it should be a number x such that 0 < x < 1.
`rnd`	integer indicating the number of decimal places to be used.

When regression = FALSE, the threshold is calculated with the Youoden's algorithm, using the pROC::coords function.

vector or matrix with performance metrics.

data(sp)

mod1 <- inla(aan ~ f(id, model = 'bym', graph = sp.adj),
             family = 'poisson', data = spn, E = eaan,
             control.predictor=list(link = 1, compute = TRUE))

mod2 <- inla(aan ~ shvn + f(id, model = 'bym', graph = sp.adj),
             family = 'poisson', data = spn, E = eaan,
             control.predictor=list(link = 1, compute = TRUE))
RMSE(mod1, mod2, observed = spn$aan / spn$eaan)