Github user mengxr commented on a diff in the pull request:
https://github.com/apache/spark/pull/1733#discussion_r16024031
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/stat/test/ChiSquaredTest.scala ---
@@ -0,0 +1,220 @@
+/*
+ * 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.test
+
+import breeze.linalg.{DenseMatrix => BDM}
+import cern.jet.stat.Probability.chiSquareComplemented
+
+import org.apache.spark.Logging
+import org.apache.spark.mllib.linalg.{Matrices, Matrix, Vector, Vectors}
+import org.apache.spark.mllib.regression.LabeledPoint
+import org.apache.spark.rdd.RDD
+
+/**
+ * Conduct the chi-squared test for the input RDDs using the specified
method.
+ * Goodness-of-fit test is conducted on two `Vectors`, whereas test of
independence is conducted
+ * on an input of type `Matrix` in which independence between columns is
assessed.
+ * We also provide a method for computing the chi-squared statistic
between each feature and the
+ * label for an input `RDD[LabeledPoint]`, return an
`Array[ChiSquaredTestResult]` of size =
+ * number of features in the inpuy RDD.
+ *
+ * Supported methods for goodness of fit: `pearson` (default)
+ * Supported methods for independence: `pearson` (default)
+ *
+ * More information on Chi-squared test:
http://en.wikipedia.org/wiki/Chi-squared_test
+ */
+private[stat] object ChiSqTest extends Logging {
+
+ /**
+ * @param name String name for the method.
+ * @param chiSqFunc Function for computing the statistic given the
observed and expected counts.
+ */
+ case class Method(name: String, chiSqFunc: (Double, Double) => Double)
+
+ // Pearson's chi-squared test:
http://en.wikipedia.org/wiki/Pearson%27s_chi-squared_test
+ val PEARSON = new Method("pearson", (observed: Double, expected: Double)
=> {
+ val dev = observed - expected
+ dev * dev / expected
+ })
+
+ // Null hypothesis for the two different types of chi-squared tests to
be included in the result.
+ object NullHypothesis extends Enumeration {
+ type NullHypothesis = Value
+ val goodnessOfFit = Value("observed follows the same distribution as
expected.")
+ val independence = Value("observations in each column are
statistically independent.")
+ }
+
+ // Method identification based on input methodName string
+ private def methodFromString(methodName: String): Method = {
+ methodName match {
+ case PEARSON.name => PEARSON
+ case _ => throw new IllegalArgumentException("Unrecognized method
for Chi squared test.")
+ }
+ }
+
+ /**
+ * Conduct Pearson's independence test for each feature against the
label across the input RDD.
+ * The contingency table is constructed from the raw (feature, label)
pairs and used to conduct
+ * the independence test.
+ * Returns an array containing the ChiSquaredTestResult for every
feature against the label.
+ */
+ def chiSquaredFeatures(data: RDD[LabeledPoint],
+ methodName: String = PEARSON.name): Array[ChiSqTestResult] = {
+ val numCols = data.first().features.size
+ val results = new Array[ChiSqTestResult](numCols)
+ var labels = Array[Double]()
+ // At most 100 columns at a time
+ val batchSize = 100
+ var batch = 0
+ while (batch * batchSize < numCols) {
+ // The following block of code can be cleaned up and made public as
+ // chiSquared(data: RDD[(V1, V2)])
+ val startCol = batch * batchSize
+ val endCol = startCol + math.min(batchSize, numCols - startCol)
+ val pairCounts = data.flatMap { p =>
+ // assume dense vectors
+ p.features.toArray.slice(startCol, endCol).zipWithIndex.map { case
(feature, col) =>
+ (col, feature, p.label)
+ }
+ }.countByValue()
+
+ if (labels.size == 0) {
+ // Do this only once for the first column since labels are
invariant across features.
+ labels = pairCounts.keys.filter(_._1 ==
startCol).map(_._3).toArray.distinct
+ }
+ val numLabels = labels.size
+ pairCounts.keys.groupBy(_._1).map { case (col, keys) =>
+ val features = keys.map(_._2).toArray.distinct
+ val numRows = features.size
+ val contingency = new BDM(numRows, numLabels, new
Array[Double](numRows * numLabels))
+ keys.foreach { case (_, feature, label) =>
+ val i = features.indexOf(feature)
+ val j = labels.indexOf(label)
+ contingency(i, j) += pairCounts((col, feature, label))
+ }
+ results(col) = chiSquaredMatrix(Matrices.fromBreeze(contingency),
methodName)
+ }
+ batch += 1
+ }
+ results
+ }
+
+ /*
+ * Pearon's goodness of fit test on the input observed and expected
counts/relative frequencies.
+ * Uniform distribution is assumed when `expected` is not passed in.
+ */
+ def chiSquared(observed: Vector,
+ expected: Vector = Vectors.dense(Array[Double]()),
+ methodName: String = PEARSON.name): ChiSqTestResult = {
+
+ // Validate input arguments
+ val method = methodFromString(methodName)
+ if (expected.size != 0 && observed.size != expected.size) {
+ throw new IllegalArgumentException("observed and expected must be of
the same size.")
+ }
+ val size = observed.size
+ // Avoid calling toArray on input vectors to avoid memory blow up
+ // (esp if size = Int.MaxValue for a SparseVector).
--- End diff --
We don't need to worry about this case. Having that many categories in
chi-square tests is not common and it is against the assumption of chi-square
test. 1000 is already very large.
---
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]
