iucnn_best_model: Select the Best Model After Model-testing

In azizka/IUCNN: Neural Networks to Approximate IUCN Red List Conservation Assessments

Select the Best Model After Model-testing

Description

Uses a data-frame of model-testing results generated with iucnn_modeltest as input, and finds the best model based on the chosen criterion.

Usage

iucnn_best_model(
  x,
  criterion = "val_acc",
  require_dropout = FALSE,
  verbose = FALSE
)

Arguments

x	a data.frame of model-testing results as produced by iucnn_modeltest.
criterion	name the criterion to rank models by (default="val_acc"). Valid options are "val_acc","val_loss","weighted_error", or "total_class_matches" (see details below): val_acc: highest validation accuracy val_loss: lowest validation loss weighted_error: lowest weighted error, e.g. an LC species misclassified as CR has a weighted error of 4-0 = 4, while an LC species misclassified as NT has a weighted error of 1-0 = 1. These error scores are summed across all validation predictions total_class_matches: picks the model that best reproduces the class distribution in the validation data. When picking this criterion it is not considered whether or not individual instances are predicted correctly, but instead it only looks at the overall class distribution in the predicted data.
require_dropout	logical (default=FALSE). If set to TRUE, the best model that contains a dropout rate of > 0 will be picked, even if other non-dropout models scored higher given the chosen criterion. Dropout models are required for certain functionalities within IUCNN, such as e.g. choosing a target accuracy when using predict_iucnn.
verbose	logical. Set to TRUE to print screen output. Default is FALSE.

Value

outputs an iucnn_model object containing all information about the best model.

Note

See vignette("Approximate_IUCN_Red_List_assessments_with_IUCNN") for a tutorial on how to run IUCNN.

Examples

## Not run: 

data("training_occ") #geographic occurrences of species with IUCN assessment
data("training_labels")# the corresponding IUCN assessments

# 1. Feature and label preparation
features <- iucnn_prepare_features(training_occ, type = "geographic") # Training features
labels <- iucnn_prepare_labels(training_labels, features) # Training labels

# Model-testing
model_testing_results <- iucnn_modeltest(features,
                                       labels,
                                       mode = 'nn-class',
                                       seed = 1234,
                                       dropout_rate = c(0.0,0.1,0.3),
                                       n_layers = c('30','40_20','50_30_10'),
                                       cv_fold = 2,
                                       init_logfile = TRUE)

# Selecting best model based on chosen criterion
best_iucnn_model <- iucnn_best_model(model_testing_results,
                                   criterion = 'val_acc',
                                   require_dropout = TRUE)

## End(Not run)

azizka/IUCNN documentation built on March 29, 2024, 9:38 a.m.