How to organize registerd models in R with mlflow?

Question

• Using aliases in mlflow is best practice and convienient to organize registered models.
• But unfortunately, even the latest mlflow R Package (Version 2.14.1) does not include any methods that allow setting aliases.
• Whereas in Python one can use client.set_registered_model_alias() or client.delete_registered_model_alias(), the R package only provides mlflow::mlflow-transition-model-version-stage at the moment.
• The concept of stages will be depricated...

Questions

• How should I deal with this situation, should I just use stages until the R package gets updated? What would be the most sensible approach here? Using tags and accepting the problem that tags are not unique?
• Does anyone know if the alias related methods will be introduced into the R package anytime soon?

Answer 1

I found a somewhat okeyish workaround: I just use the mlflow rest api (https://mlflow.org/docs/latest/rest-api.html#set-registered-model-alias) from R to set the alias...

mlflowSetRegisteredModelAlias <- function(mlflowUri, modelName, alias, version) {
  # Define api request payload
  payload <- list(name = modelName,
                  alias = alias,
                  version = as.character(version))
  
  # Make the POST request
  mlflowApiEndpoint <- paste0(mlflowUri, "api/2.0/mlflow/registered-models/alias")
  response <- mlflowApiEndpoint %>%
    httr2::request() %>%
    httr2::req_body_json(payload) %>%
    httr2::req_perform()
}

For deleting an alias it would be with ad DELETE method:

 payload <- list(name = modelName, alias = alias)
  
  # Make the DELETE request
  response <-  mlflowApiEndpoint %>%
    httr2::request() %>%
    httr2::req_method("DELETE") %>%
    httr2::req_body_json(payload) %>%
    httr2::req_perform()