Github user zapletal-martin commented on a diff in the pull request:
https://github.com/apache/spark/pull/3519#discussion_r23841288
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/regression/IsotonicRegression.scala
---
@@ -0,0 +1,238 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.mllib.regression
+
+import java.io.Serializable
+import java.util.Arrays.binarySearch
+
+import org.apache.spark.api.java.{JavaDoubleRDD, JavaRDD}
+import org.apache.spark.rdd.RDD
+
+/**
+ * Regression model for Isotonic regression
+ *
+ * @param features Array of features.
+ * @param labels Array of labels associated to the features at the same
index.
+ */
+class IsotonicRegressionModel (
+ features: Array[Double],
+ val labels: Array[Double])
+ extends Serializable {
+
+ /**
+ * Predict labels for provided features
+ * Using a piecewise constant function
+ *
+ * @param testData features to be labeled
+ * @return predicted labels
+ */
+ def predict(testData: RDD[Double]): RDD[Double] =
+ testData.map(predict)
+
+ /**
+ * Predict labels for provided features
+ * Using a piecewise constant function
+ *
+ * @param testData features to be labeled
+ * @return predicted labels
+ */
+ def predict(testData: JavaRDD[java.lang.Double]): JavaDoubleRDD =
+ JavaDoubleRDD.fromRDD(predict(testData.rdd.asInstanceOf[RDD[Double]]))
+
+ /**
+ * Predict a single label
+ * Using a piecewise constant function
+ *
+ * @param testData feature to be labeled
+ * @return predicted label
+ */
+ def predict(testData: Double): Double = {
+ val result = binarySearch(features, testData)
+
+ val index =
+ if (result == -1) {
--- End diff --
As for the special singularity case I believe this requires further
considerations. Currently we just sort the input to PAV by feature therefore
order of multiple data points with the same feature is undefined.
Consider a case where features are 1, 2, 2, 3 and labels are in first case
1, 4, 2, 5 and in second case 1, 2, 4, 5. For first case the result of PAV
would be 1, 3, 3, 5 but in second case 1, 2, 4, 5.
Similarly for `IsotonicRegressionModel` with boundaries 1, 2, 2, 3 and
predictions in first case 1, 4, 2, 5 and in second case 1, 2, 4, 5. Now the
first mode would return predict(1.5)=2.5, predict(2.5)=3.5, but the second
would return 1.5 and 4.5 respectively for the same input values.
I suggest to sort the input by features and then by labels if features are
equal. The same would be true for the model. Therefore both PAV and the
predictions of values between boundaries would be deterministic. The
predictions for the boundary with multiple values would remain
non-deterministic (based on `Java.util.Arrays.binarySearch()` which in this
case also returns one of the correct results, but does not specify which).
---
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]
