featureImportances method doesn't exist

[gnani4444]

Hi
I am using XGBoost Spark 3.0 GPU version

I couldn't find featureImportances method for the model object. Can you guide me how to get feature importances from the trained model.
and
Can you share any notebook or code for hyper-parameter tuning using hyperopt, If you already have it.

Thanks in advance

[wbo4958]

the pyspark supporting for XGBoost is totally different from XGBoost-built-in-python package. Actually xgboost pyspark is just a wrapper of XGBoost4j. So there is no such method.

But I suppose what you're looking for is

model.nativeBooster.getScore("xxx", "xxx")

  /**
    * Get importance of each feature based on information gain or cover
    * Supported: ["gain, "cover", "total_gain", "total_cover"]
    *
    * @return featureScoreMap  key: feature index, value: feature importance score
    */
  @throws(classOf[XGBoostError])
  def getScore(featureMap: String, importanceType: String): Map[String, Double] = {
    Map(booster.getScore(featureMap, importanceType)
        .asScala.mapValues(_.doubleValue).toSeq: _*)
  }

  /**
    * Get importance of each feature based on information gain or cover
    * , with specified feature names.
    * Supported: ["gain, "cover", "total_gain", "total_cover"]
    *
    * @return featureScoreMap  key: feature name, value: feature importance score
    */
  @throws(classOf[XGBoostError])
  def getScore(featureNames: Array[String], importanceType: String): Map[String, Double] = {
    Map(booster.getScore(featureNames, importanceType)
        .asScala.mapValues(_.doubleValue).toSeq: _*)
  }

[wbo4958]

We don't have example about hyperopt, but we have some notebooks for CrossValidator

https://github.com/NVIDIA/spark-xgboost-examples/blob/spark-3/examples/notebooks/python/cv-mortgage-gpu.ipynb

[gnani4444]

the pyspark supporting for XGBoost is totally different from XGBoost-built-in-python package. Actually xgboost pyspark is just a wrapper of XGBoost4j. So there is no such method.

But I suppose what you're looking for is

model.nativeBooster.getScore("xxx", "xxx")

  /**
    * Get importance of each feature based on information gain or cover

I got an error Method doesn't exit

[wbo4958]

can you try

model.nativeBooster.getScore("", "gain")

[gnani4444]

@wbo4958
I got a java object

I added a feature name got error

[sdev2030]

@gnani4444
You can assign the object to a variable and print it. Also you can convert the object to java list using toList() method on that object. Once you get the list, extract the index and score for each feature by looping thru the list and creating pandas data frame that can printed as a graph. Hope this helps.

[wbo4958]

@gnani4444, still has any issue?