Cloud not load xgboost model into spring boot service
hi, I have serialized a xgboost-spark pipeline model in bundle format, but I could not load it into spring boot service because of "xgboost.classifier" op is not supported. I found that mleap-xgboost-runtime is released now. It seems loading xgboost-spark pipeline should be accessible. Would you help me about how to add dependency of mleap-xgboost-runtime into mleap-spring-boot ? Or update the docker with xgboost model support?
I think this should be a matter of adding mleap-xgboost-runtime into your jvm dependency management system? I.e., add it as a dependency in your pom.xml, build.sbt, or build.gradle file depending on which build system you are using.
hi, I have add it as a dependency in our build.sbt, but it does not work, project mleap-runtime build.sbt:
import ml.combust.mleap.{Dependencies, Common} Common.defaultMleapSettings Common.defaultMleapXgboostSparkSettings Dependencies.runtime(scalaVersion)
The error is “Exception in thread "main" java.util.NoSuchElementException: key not found: xgboost.classifier at scala.collection.MapLike.default(MapLike.scala:235)”
hi, I have add it as a dependency in our build.sbt, but it does not work, project mleap-runtime build.sbt:
import ml.combust.mleap.{Dependencies, Common} Common.defaultMleapSettings Common.defaultMleapXgboostSparkSettings Dependencies.runtime(scalaVersion)
The error is “Exception in thread "main" java.util.NoSuchElementException: key not found: xgboost.classifier at scala.collection.MapLike.default(MapLike.scala:235)”
Can you provide a more complete stacktrace?
My only thought right now is that the mleap runtime can't find the correct op used to load the bundle. This could occur if you have a resources.conf file in your project and aren't having mleap's reference.conf appended to your project's reference.conf. E.g., if you are using sbt-assembly plugin, then you might need
assemblyMergeStrategy in assembly := {
case "reference.conf" => MergeStrategy.concat
case _ => MergeStrategy.first
}
But a complete stacktrace will help to confirm/deny this.