Github user mpjlu commented on a diff in the pull request:
https://github.com/apache/spark/pull/14597#discussion_r74740698
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/feature/ChiSqSelector.scala ---
@@ -197,3 +197,28 @@ class ChiSqSelector @Since("1.3.0") (
new ChiSqSelectorModel(indices)
}
}
+
+/**
+ * Creates a ChiSquared feature selector by False Positive Rate (FPR) test.
+ * @param alpha the highest p-value for features to be kept
+ */
+@Since("2.1.0")
+class ChiSqSelectorByFpr @Since("2.1.0") (
+ @Since("2.1.0") val alpha: Double) extends Serializable {
+
+ /**
+ * Returns a ChiSquared feature selector by FPR.
+ *
+ * @param data an `RDD[LabeledPoint]` containing the labeled dataset
with categorical features.
+ * Real-valued features will be treated as categorical for
each distinct value.
+ * Apply feature discretizer before using this function.
+ */
+ @Since("2.1.0")
+ def fit(data: RDD[LabeledPoint]): ChiSqSelectorModel = {
+ val indices = Statistics.chiSqTest(data)
+ .zipWithIndex.filter { case (res, _) => res.pValue < alpha }
--- End diff --
Thanks, this makes sense. It is better to add the parameter controlling
what type of feature selection is done to ChiSqSelector. But my idea is the
the selection is done in "fit" function to be consistent with other feature
selection methods, we don't need to pass what type of selection to the model.
If we consider the user can get the p-value, maybe return the p-value from
ChiSqSelector is ok? If so , we don't need to make any change to the model
class. How do you think about it?
---
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]
