Export Model to ONNX Format — export

Exports the estimator component of an mlr3 learner to ONNX format for cross-language deployment. Supports xgboost and glmnet learners.

Usage

export_onnx(learner, path, task, export_preprocessing = TRUE)

Arguments

learner: A trained mlr3 Learner or GraphLearner
path: File path for the ONNX model (with .onnx extension)
task: The mlr3 Task used for training (for metadata)
export_preprocessing: Logical, whether to also export preprocessing manifest

Value

List with export status and paths to exported files

Details

ONNX export is currently supported for: - xgboost: via native xgboost save + Python conversion (most reliable) - glmnet: via coefficient extraction (linear models)

For xgboost, this function exports the native .xgb format. A separate Python script can convert to ONNX. For glmnet, coefficients are exported as JSON.

For other learners, only the preprocessing manifest is exported. Use vetiver for R-based deployment of these models.

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("jsonlite", quietly = TRUE) &&
    requireNamespace("mlr3learners", quietly = TRUE) &&
    requireNamespace("xgboost", quietly = TRUE)) {
  # Train xgboost model
  task <- tsk("iris")
  learner <- lrn("classif.xgboost", nrounds = 50)
  learner$train(task)

  # Export to ONNX-compatible format
  export_onnx(learner, "model.onnx", task)
}
} # }