caretSDM (version 1.1.0.1)

predict_sdm: Predict SDM models in new data

Description

This function projects SDM models to new scenarios

Usage

predict_sdm(m,
            scen = NULL,
            metric = "ROC",
            th = 0.9,
            tp = "prob",
            ensembles = TRUE,
            file = NULL,
            add.current = TRUE)
get_predictions(i)
get_ensembles(i)

Value

A input_sdm or a predictions object.

Arguments

m

A input_sdm or a models object.

scen

A scenarios object or NULL. If NULL and m is a input_sdm with a scenarios slot, it will be used.

metric

A character containing the metric in which the th will be calculated/applied. Default is ROC. See ?mean_validation_metrics for the metrics available.

th

Thresholds for metrics. Can be numeric or a function.

tp

Type of output to be retrieved. See details.

ensembles

Boolean. Should ensembles be calculated? If TRUE a series of ensembles are obtained. See details.

file

File to sabe predictions.

add.current

If current scenario is not available, predictors will be used as the current scenario.

i

A input_sdm or a predictions object.

Author

Luíz Fernando Esser ([e-mail]) https://luizfesser.wordpress.com

Details

tp is a parameter to be passed on caret to retrieve either the probabilities of classes (tp="prob") or the raw output (tp="raw"), which could vary depending on the algorithm used, but usually would be on of the classes (factor vector with presences and pseudoabsences).

When ensembles is set to TRUE, three ensembles are currently implemented. mean_occ_prob is the mean occurrence probability, which is a simple mean of predictions, wmean_AUC is the same mean_occ_prob, but weighted by AUC, and committee_avg is the committee average, as known as majority rule, where predictions are binarized and then a mean is obtained.

get_predictions returns the list of all predictions to all scenarios, all species, all algorithms and all repetitions. Useful for those who wish to implement their own ensemble methods.

get_ensembles returns a matrix of data.frames, where each column is a scenario and each row is a species.

scenarios_names returns the scenarios names in a sdm_area or input_sdm object.

get_scenarios_data returns the data from scenarios in a sdm_area or input_sdm object.

See Also

sdm_area input_sdm mean_validation_metrics

Examples

# Create sdm_area object:
set.seed(1)
sa <- sdm_area(parana, cell_size = 100000, crs = 6933)

# Include predictors:
sa <- add_predictors(sa, bioc) |> select_predictors(c("bio1", "bio12"))

# Include scenarios:
sa <- add_scenarios(sa)

# Create occurrences:
oc <- occurrences_sdm(occ, crs = 6933) |> join_area(sa)

# Create input_sdm:
i <- input_sdm(oc, sa)

# Pseudoabsence generation:
i <- pseudoabsences(i, method="random", n_set=2)

# Custom trainControl:
ctrl_sdm <- caret::trainControl(method = "boot",
                                number = 1,
                                repeats = 1,
                                classProbs = TRUE,
                                returnResamp = "all",
                                summaryFunction = summary_sdm,
                                savePredictions = "all")

# Train models:
i <- train_sdm(i, algo = c("naive_bayes"), ctrl=ctrl_sdm) |>
  suppressWarnings()

# Predict models:
i  <- predict_sdm(i, th = 0.8)
i