Github user sethah commented on a diff in the pull request:
https://github.com/apache/spark/pull/16441#discussion_r94861226
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/classification/GBTClassifier.scala ---
@@ -248,12 +269,38 @@ class GBTClassificationModel private[ml](
if (prediction > 0.0) 1.0 else 0.0
}
+ override protected def predictRaw(features: Vector): Vector = {
+ val treePredictions =
_trees.map(_.rootNode.predictImpl(features).prediction)
+ val prediction = blas.ddot(numTrees, treePredictions, 1, _treeWeights,
1)
+ Vectors.dense(Array(-prediction, prediction))
+ }
+
+ override protected def raw2probabilityInPlace(rawPrediction: Vector):
Vector = {
+ rawPrediction match {
+ // The probability can be calculated for positive result:
+ // p+(x) = 1 / (1 + e^(-2 * F(x)))
+ // and negative result:
+ // p-(x) = 1 / (1 + e^(2 * F(x)))
+ case dv: DenseVector =>
+ var i = 0
+ val size = dv.size
+ while (i < size) {
+ dv.values(i) = 1 / (1 + math.exp(-2 * dv.values(i)))
--- End diff --
my concern is that this is hard coded to logistic loss. Maybe we can add a
static method to GBTClassificationModel
````scala
private def classProbability(class: Int, loss: String, rawPrediction:
Double): Double = {
loss match {
case "logistic" => ...
case _ => throw new Exception("Only logistic loss is supported ...")
}
}
````
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
