Github user mengxr commented on a diff in the pull request:
https://github.com/apache/spark/pull/1367#discussion_r14842782
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/stat/correlation/Correlation.scala
---
@@ -0,0 +1,121 @@
+/*
+ * 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.stat.correlation
+
+import org.apache.spark.mllib.linalg.{DenseVector, Matrix, Vector}
+import org.apache.spark.rdd.RDD
+
+/**
+ * New correlation algorithms should implement this trait
+ */
+trait Correlation {
+
+ /**
+ * Compute correlation for two datasets.
+ */
+ def computeCorrelation(x: RDD[Double], y: RDD[Double]): Double
+
+ /**
+ * Compute the correlation matrix S, for the input matrix, where S(i, j)
is the correlation
+ * between column i and j.
+ */
+ def computeCorrelationMatrix(X: RDD[Vector]): Matrix
+
+ /**
+ * Combine the two input RDD[Double]s into an RDD[Vector] and compute
the correlation using the
+ * correlation implementation for RDD[Vector]
+ */
+ def computeCorrelationWithMatrixImpl(x: RDD[Double], y: RDD[Double]):
Double = {
+ val mat: RDD[Vector] = x.zip(y).mapPartitions({ iter =>
+ iter.map {case(xi, yi) => new DenseVector(Array(xi, yi))}
+ }, preservesPartitioning = true)
+ computeCorrelationMatrix(mat)(0, 1)
+ }
+
+}
+
+/**
+ * Delegates computation to the specific correlation object based on the
input method name
+ *
+ * Currently supported correlations: pearson, spearman.
+ * After new correlation algorithms are added, please update the
documentation here and in
+ * Statistics.scala for the correlation APIs.
+ *
+ * Cases are ignored when doing method matching. We also allow initials,
e.g. "P" for "pearson", as
+ * long as initials are unique in the supported set of correlation
algorithms. In addition, a
--- End diff --
R only supports 3 correlations and they happen to not have name collisions,
which is fine because there was no plan to support more correlations in R on a
single machine. But if there comes a popular correlation called "kuma", the way
R can handle it is renaming it to some name like "zuma" or explaining to user
why "ku" is required for "kuma" but "k" is for "kendall". That was a bad
decision R's developers made.
The implementation in this PR suggests that developers can add their own
implementation but the name resolution isn't really expandable. Besides
kendall's tau, we might add approximate algorithms in the future, which could
have name like "approx_spearman" and "approx_pearson". Then we are facing
collisions in the first character or non-unique substring matching.
Also, we want to make the behavior consistent in MLlib. It adds burden to
future development. The best solution to me is exact string matching, simple
implementation and low maintenance cost.
---
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.
---
