Save a Trained Pixel Model

Saves a complete pixel classifier as one model-bundle directory. The bundle contains the trained Keras network and the R metadata needed by apply_pixel_model, including class names, input dimensions, and the fitted land-cover codes. Move or archive the complete folder rather than the single files inside it.

Usage

save_pixel_model(model, path, overwrite = FALSE)

Arguments

model

List. Trained pixel model returned by train_pixel_model.

path

Character. Directory in which to save the complete model bundle. The path must not end in .keras or .rds; those files are managed inside the directory.

overwrite

Logical. Whether to replace the model files in an existing directory (default: FALSE). Other files in that directory are not removed.

Value

The normalized bundle path, invisibly.

See also

load_pixel_model, train_pixel_model

Other neural network training: set_random_seed(), train_metric_model(), train_pixel_model()

Examples

if (FALSE) { # requireNamespace("reticulate", quietly = TRUE) && reticulate::virtualenv_exists("r-keras")
training_landscapes <- create_landscapes(
  n = 6,
  patterns = c("sharp", "random"),
  width = 20,
  height = 20
)
set_random_seed(42)
model <- train_pixel_model(
  training_landscapes,
  cv_method = "none",
  epochs = 1,
  verbose = FALSE
)

model_bundle <- tempfile("pixel-model-")
save_pixel_model(model, model_bundle)
reloaded_model <- load_pixel_model(model_bundle)
# Clean up the temporary bundle created for this example.
unlink(model_bundle, recursive = TRUE)
}