Within the TidySDM package, you can define what models you wish to include in your ensemble with the "workflow_set" function. When including a maxent model you can use the argument "sdm_spec_maxent" and set the "tune" parameter to sdm, all, custom, or none. I wish to use the custom tune setting and manually set the feature class and regularization multiplier options. I wish to provide a range of values which I have based on the help file Maxent_params. However I receive an error each time that says, "the condition has length > 1".
Here is the exact code I am using for the workflow set, as based on the tidySDM github overview example and the help files.
dive_models <- workflow_set(
preproc = list(default = dive.rec),
models = list(
glm = sdm_spec_glm(),
gbm = sdm_spec_boost_tree(
tune = "sdm"
),
maxent = maxent(
mode = "classification",
engine = "maxnet",
feature_classes(values = c("L","Q", "H", "T", "LQ", "LH", "LT", "QH", "QT", "HT", "LQH", "LQT", "QHT", "LHT", "LQHT")),
regularization_multiplier(range = c(0.5, 3), trans = NULL)
),
mars = mars(
mode = "classification",
engine = "earth",
prod_degree = 4
)
),
cross = T
) |>
option_add(control = control_ensemble_grid())
Does anyone see what i may be missing since I have formatted this exactly as it is in the help file?
To use a custom range for tuning parameters, you need to create a custom object with parameters. Here is a reprex changing the two parameters for maxent:
library(tidysdm)
#> Loading required package: tidymodels
#> Loading required package: spatialsample
lacerta_thin <- terra::readRDS(system.file("extdata/lacerta_thin_all_vars.rds",
package = "tidysdm"))
# create a recipe
lacerta_rec <- recipe(lacerta_thin, formula = class ~ .)
# set up custom parameters ranges for maxent
maxent_param <- parameters(regularization_multiplier(),feature_classes()) %>%
update(regularization_multiplier = regularization_multiplier(c(0.1,0.5))) %>%
update(feature_classes = feature_classes(c("l","h")))
# create the workflow set
lacerta_models <-
workflow_set(
preproc = list(default = lacerta_rec),
models = list(
# the standard glm specs
glm = sdm_spec_glm(),
# maxent specs with tuning
maxent = maxent(
mode = "classification",
engine = "maxnet",
regularization_multiplier = tune(),
feature_classes = tune()
)
),
# make all combinations of preproc and models,
cross = TRUE
) %>%
# tweak controls to store information needed later to create the ensemble
option_add(control = control_ensemble_grid()) %>%
# add the custom maxent parameter ranges
option_add(param_info = maxent_param,id="default_maxent")
# spatial block
lacerta_cv <- spatial_block_cv(data = lacerta_thin, v = 3, n = 5)
# tune the models
set.seed(1234567)
lacerta_models <-
lacerta_models %>%
workflow_map("tune_grid",
resamples = lacerta_cv, grid = 3,
metrics = sdm_metric_set(), verbose = TRUE
)
#> i No tuning parameters. `fit_resamples()` will be attempted
#> i 1 of 2 resampling: default_glm
#> → A | warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
#> There were issues with some computations A: x1
#> There were issues with some computations A: x1
#>
#> ✔ 1 of 2 resampling: default_glm (128ms)
#> i 2 of 2 tuning: default_maxent
#> ✔ 2 of 2 tuning: default_maxent (1.2s)
# collect the results for the maxent model
workflowsets::extract_workflow_set_result(lacerta_models,id="default_maxent") %>% collect_metrics()
#> # A tibble: 9 × 8
#> regularization_multip…¹ feature_classes .metric .estimator mean n std_err
#> <dbl> <chr> <chr> <chr> <dbl> <int> <dbl>
#> 1 0.230 h boyce_… binary 0.317 3 0.269
#> 2 0.230 h roc_auc binary 0.812 3 0.0356
#> 3 0.230 h tss_max binary 0.589 3 0.0513
#> 4 0.252 l boyce_… binary 0.533 3 0.139
#> 5 0.252 l roc_auc binary 0.817 3 0.0421
#> 6 0.252 l tss_max binary 0.594 3 0.0865
#> 7 0.451 l boyce_… binary 0.614 3 0.0578
#> 8 0.451 l roc_auc binary 0.815 3 0.0423
#> 9 0.451 l tss_max binary 0.601 3 0.0937
#> # ℹ abbreviated name: ¹regularization_multiplier
#> # ℹ 1 more variable: .config <chr>
Created on 2025-02-13 with reprex v2.1.1