You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@spark.apache.org by db...@apache.org on 2016/10/14 20:21:13 UTC
spark git commit: [SPARK-17941][ML][TEST] Logistic regression tests
should use sample weights.
Repository: spark
Updated Branches:
refs/heads/master 05800b4b4 -> de1c1ca5c
[SPARK-17941][ML][TEST] Logistic regression tests should use sample weights.
## What changes were proposed in this pull request?
The sample weight testing for logistic regressions is not robust. Logistic regression suite already has many test cases comparing results to R glmnet. Since both libraries support sample weights, we should use sample weights in the test to increase coverage for sample weighting. This patch doesn't really add any code and makes the testing more complete.
Also fixed some errors with the R code that was referenced in the test suit. Changed `standardization=T` to `standardize=T` since the former is invalid.
## How was this patch tested?
Existing unit tests are modified. No non-test code is touched.
Author: sethah <se...@gmail.com>
Closes #15488 from sethah/logreg_weight_tests.
Project: http://git-wip-us.apache.org/repos/asf/spark/repo
Commit: http://git-wip-us.apache.org/repos/asf/spark/commit/de1c1ca5
Tree: http://git-wip-us.apache.org/repos/asf/spark/tree/de1c1ca5
Diff: http://git-wip-us.apache.org/repos/asf/spark/diff/de1c1ca5
Branch: refs/heads/master
Commit: de1c1ca5c9d6064d3b7b3711e3bfb08fa018abe8
Parents: 05800b4
Author: sethah <se...@gmail.com>
Authored: Fri Oct 14 20:21:03 2016 +0000
Committer: DB Tsai <db...@login.dbtsai.com>
Committed: Fri Oct 14 20:21:03 2016 +0000
----------------------------------------------------------------------
.../LogisticRegressionSuite.scala | 1493 +++++++++---------
1 file changed, 748 insertions(+), 745 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/spark/blob/de1c1ca5/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
----------------------------------------------------------------------
diff --git a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
index 42b5675..bc631dc 100644
--- a/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/ml/classification/LogisticRegressionSuite.scala
@@ -25,14 +25,14 @@ import scala.util.control.Breaks._
import org.apache.spark.{SparkException, SparkFunSuite}
import org.apache.spark.ml.attribute.NominalAttribute
import org.apache.spark.ml.classification.LogisticRegressionSuite._
-import org.apache.spark.ml.feature.LabeledPoint
+import org.apache.spark.ml.feature.{Instance, LabeledPoint}
import org.apache.spark.ml.linalg.{DenseMatrix, Matrices, SparseMatrix, SparseVector, Vector, Vectors}
import org.apache.spark.ml.param.ParamsSuite
import org.apache.spark.ml.util.{DefaultReadWriteTest, MLTestingUtils}
import org.apache.spark.ml.util.TestingUtils._
import org.apache.spark.mllib.util.MLlibTestSparkContext
import org.apache.spark.sql.{Dataset, Row}
-import org.apache.spark.sql.functions.{col, lit}
+import org.apache.spark.sql.functions.{col, lit, rand}
import org.apache.spark.sql.types.LongType
class LogisticRegressionSuite
@@ -40,6 +40,7 @@ class LogisticRegressionSuite
import testImplicits._
+ private val seed = 42
@transient var smallBinaryDataset: Dataset[_] = _
@transient var smallMultinomialDataset: Dataset[_] = _
@transient var binaryDataset: Dataset[_] = _
@@ -49,7 +50,7 @@ class LogisticRegressionSuite
override def beforeAll(): Unit = {
super.beforeAll()
- smallBinaryDataset = generateLogisticInput(1.0, 1.0, nPoints = 100, seed = 42).toDF()
+ smallBinaryDataset = generateLogisticInput(1.0, 1.0, nPoints = 100, seed = seed).toDF()
smallMultinomialDataset = {
val nPoints = 100
@@ -61,7 +62,7 @@ class LogisticRegressionSuite
val xVariance = Array(0.6856, 0.1899)
val testData = generateMultinomialLogisticInput(
- coefficients, xMean, xVariance, addIntercept = true, nPoints, 42)
+ coefficients, xMean, xVariance, addIntercept = true, nPoints, seed)
val df = sc.parallelize(testData, 4).toDF()
df.cache()
@@ -76,9 +77,9 @@ class LogisticRegressionSuite
val testData =
generateMultinomialLogisticInput(coefficients, xMean, xVariance,
- addIntercept = true, nPoints, 42)
+ addIntercept = true, nPoints, seed)
- sc.parallelize(testData, 4).toDF()
+ sc.parallelize(testData, 4).toDF().withColumn("weight", rand(seed))
}
multinomialDataset = {
@@ -91,9 +92,9 @@ class LogisticRegressionSuite
val xVariance = Array(0.6856, 0.1899, 3.116, 0.581)
val testData = generateMultinomialLogisticInput(
- coefficients, xMean, xVariance, addIntercept = true, nPoints, 42)
+ coefficients, xMean, xVariance, addIntercept = true, nPoints, seed)
- val df = sc.parallelize(testData, 4).toDF()
+ val df = sc.parallelize(testData, 4).toDF().withColumn("weight", rand(seed))
df.cache()
df
}
@@ -104,11 +105,11 @@ class LogisticRegressionSuite
* so we can validate the training accuracy compared with R's glmnet package.
*/
ignore("export test data into CSV format") {
- binaryDataset.rdd.map { case Row(label: Double, features: Vector) =>
- label + "," + features.toArray.mkString(",")
+ binaryDataset.rdd.map { case Row(label: Double, features: Vector, weight: Double) =>
+ label + "," + weight + "," + features.toArray.mkString(",")
}.repartition(1).saveAsTextFile("target/tmp/LogisticRegressionSuite/binaryDataset")
- multinomialDataset.rdd.map { case Row(label: Double, features: Vector) =>
- label + "," + features.toArray.mkString(",")
+ multinomialDataset.rdd.map { case Row(label: Double, features: Vector, weight: Double) =>
+ label + "," + weight + "," + features.toArray.mkString(",")
}.repartition(1).saveAsTextFile("target/tmp/LogisticRegressionSuite/multinomialDataset")
}
@@ -519,31 +520,35 @@ class LogisticRegressionSuite
test("binary logistic regression with intercept without regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(true).setStandardization(true)
+ .setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true).setStandardization(false)
+ .setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
-
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0, lambda = 0))
- coefficients
+ Use the following R code to load the data and train the model using glmnet package.
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0,
+ lambda = 0))
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) 2.7355261
+ data.V3 -0.5734389
+ data.V4 0.8911736
+ data.V5 -0.3878645
+ data.V6 -0.8060570
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) 2.8366423
- data.V2 -0.5895848
- data.V3 0.8931147
- data.V4 -0.3925051
- data.V5 -0.7996864
*/
- val interceptR = 2.8366423
- val coefficientsR = Vectors.dense(-0.5895848, 0.8931147, -0.3925051, -0.7996864)
+ val coefficientsR = Vectors.dense(-0.5734389, 0.8911736, -0.3878645, -0.8060570)
+ val interceptR = 2.7355261
assert(model1.intercept ~== interceptR relTol 1E-3)
assert(model1.coefficients ~= coefficientsR relTol 1E-3)
@@ -555,413 +560,374 @@ class LogisticRegressionSuite
test("binary logistic regression without intercept without regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false).setStandardization(true)
+ .setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false).setStandardization(false)
+ .setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
+ Use the following R code to load the data and train the model using glmnet package.
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients =
- coef(glmnet(features,label, family="binomial", alpha = 0, lambda = 0, intercept=FALSE))
- coefficients
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0,
+ lambda = 0, intercept=FALSE))
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) .
+ data.V3 -0.3448461
+ data.V4 1.2776453
+ data.V5 -0.3539178
+ data.V6 -0.7469384
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) .
- data.V2 -0.3534996
- data.V3 1.2964482
- data.V4 -0.3571741
- data.V5 -0.7407946
*/
- val interceptR = 0.0
- val coefficientsR = Vectors.dense(-0.3534996, 1.2964482, -0.3571741, -0.7407946)
+ val coefficientsR = Vectors.dense(-0.3448461, 1.2776453, -0.3539178, -0.7469384)
- assert(model1.intercept ~== interceptR relTol 1E-3)
+ assert(model1.intercept ~== 0.0 relTol 1E-3)
assert(model1.coefficients ~= coefficientsR relTol 1E-2)
// Without regularization, with or without standardization should converge to the same solution.
- assert(model2.intercept ~== interceptR relTol 1E-3)
+ assert(model2.intercept ~== 0.0 relTol 1E-3)
assert(model2.coefficients ~= coefficientsR relTol 1E-2)
}
test("binary logistic regression with intercept with L1 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(true)
+ .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(false)
+ .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
+ Use the following R code to load the data and train the model using glmnet package.
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 1, lambda = 0.12))
- coefficients
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 1,
+ lambda = 0.12, standardize=T))
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) -0.06775980
+ data.V3 .
+ data.V4 .
+ data.V5 -0.03933146
+ data.V6 -0.03047580
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) -0.05627428
- data.V2 .
- data.V3 .
- data.V4 -0.04325749
- data.V5 -0.02481551
*/
- val interceptR1 = -0.05627428
- val coefficientsR1 = Vectors.dense(0.0, 0.0, -0.04325749, -0.02481551)
+ val coefficientsRStd = Vectors.dense(0.0, 0.0, -0.03933146, -0.03047580)
+ val interceptRStd = -0.06775980
- assert(model1.intercept ~== interceptR1 relTol 1E-2)
- assert(model1.coefficients ~= coefficientsR1 absTol 2E-2)
+ assert(model1.intercept ~== interceptRStd relTol 1E-2)
+ assert(model1.coefficients ~= coefficientsRStd absTol 2E-2)
/*
- Using the following R code to load the data and train the model using glmnet package.
+ Use the following R code to load the data and train the model using glmnet package.
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 1, lambda = 0.12,
- standardize=FALSE))
- coefficients
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 1,
+ lambda = 0.12, standardize=F))
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) 0.3544768
+ data.V3 .
+ data.V4 .
+ data.V5 -0.1626191
+ data.V6 .
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) 0.3722152
- data.V2 .
- data.V3 .
- data.V4 -0.1665453
- data.V5 .
*/
- val interceptR2 = 0.3722152
- val coefficientsR2 = Vectors.dense(0.0, 0.0, -0.1665453, 0.0)
+ val coefficientsR = Vectors.dense(0.0, 0.0, -0.1626191, 0.0)
+ val interceptR = 0.3544768
- assert(model2.intercept ~== interceptR2 relTol 1E-2)
- assert(model2.coefficients ~== coefficientsR2 absTol 1E-3)
+ assert(model2.intercept ~== interceptR relTol 1E-2)
+ assert(model2.coefficients ~== coefficientsR absTol 1E-3)
// TODO: move this to a standalone test of compression after SPARK-17471
assert(model2.coefficients.isInstanceOf[SparseVector])
}
test("binary logistic regression without intercept with L1 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(true)
+ .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(false)
+ .setElasticNetParam(1.0).setRegParam(0.12).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
-
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 1, lambda = 0.12,
- intercept=FALSE))
- coefficients
-
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) .
- data.V2 .
- data.V3 .
- data.V4 -0.05189203
- data.V5 -0.03891782
- */
- val interceptR1 = 0.0
- val coefficientsR1 = Vectors.dense(0.0, 0.0, -0.05189203, -0.03891782)
-
- assert(model1.intercept ~== interceptR1 relTol 1E-3)
- assert(model1.coefficients ~= coefficientsR1 absTol 1E-3)
+ Use the following R code to load the data and train the model using glmnet package.
- /*
- Using the following R code to load the data and train the model using glmnet package.
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="binomial", alpha = 1,
+ lambda = 0.12, intercept=F, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 1,
+ lambda = 0.12, intercept=F, standardize=F))
+ coefficientsStd
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) .
+ data.V3 .
+ data.V4 .
+ data.V5 -0.04967635
+ data.V6 -0.04757757
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 1, lambda = 0.12,
- intercept=FALSE, standardize=FALSE))
- coefficients
+ coefficients
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) .
+ data.V3 .
+ data.V4 .
+ data.V5 -0.08433195
+ data.V6 .
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) .
- data.V2 .
- data.V3 .
- data.V4 -0.08420782
- data.V5 .
*/
- val interceptR2 = 0.0
- val coefficientsR2 = Vectors.dense(0.0, 0.0, -0.08420782, 0.0)
+ val coefficientsRStd = Vectors.dense(0.0, 0.0, -0.04967635, -0.04757757)
- assert(model2.intercept ~== interceptR2 absTol 1E-3)
- assert(model2.coefficients ~= coefficientsR2 absTol 1E-3)
+ val coefficientsR = Vectors.dense(0.0, 0.0, -0.08433195, 0.0)
+
+ assert(model1.intercept ~== 0.0 absTol 1E-3)
+ assert(model1.coefficients ~= coefficientsRStd absTol 1E-3)
+ assert(model2.intercept ~== 0.0 absTol 1E-3)
+ assert(model2.coefficients ~= coefficientsR absTol 1E-3)
}
test("binary logistic regression with intercept with L2 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(true)
+ .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(false)
+ .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
-
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0, lambda = 1.37))
- coefficients
-
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) 0.15021751
- data.V2 -0.07251837
- data.V3 0.10724191
- data.V4 -0.04865309
- data.V5 -0.10062872
- */
- val interceptR1 = 0.15021751
- val coefficientsR1 = Vectors.dense(-0.07251837, 0.10724191, -0.04865309, -0.10062872)
-
- assert(model1.intercept ~== interceptR1 relTol 1E-3)
- assert(model1.coefficients ~= coefficientsR1 relTol 1E-3)
+ Use the following R code to load the data and train the model using glmnet package.
- /*
- Using the following R code to load the data and train the model using glmnet package.
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0,
+ lambda = 1.37, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0,
+ lambda = 1.37, standardize=F))
+ coefficientsStd
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) 0.12707703
+ data.V3 -0.06980967
+ data.V4 0.10803933
+ data.V5 -0.04800404
+ data.V6 -0.10165096
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0, lambda = 1.37,
- standardize=FALSE))
- coefficients
+ coefficients
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) 0.46613016
+ data.V3 -0.04944529
+ data.V4 0.02326772
+ data.V5 -0.11362772
+ data.V6 -0.06312848
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) 0.48657516
- data.V2 -0.05155371
- data.V3 0.02301057
- data.V4 -0.11482896
- data.V5 -0.06266838
*/
- val interceptR2 = 0.48657516
- val coefficientsR2 = Vectors.dense(-0.05155371, 0.02301057, -0.11482896, -0.06266838)
+ val coefficientsRStd = Vectors.dense(-0.06980967, 0.10803933, -0.04800404, -0.10165096)
+ val interceptRStd = 0.12707703
+ val coefficientsR = Vectors.dense(-0.04944529, 0.02326772, -0.11362772, -0.06312848)
+ val interceptR = 0.46613016
- assert(model2.intercept ~== interceptR2 relTol 1E-3)
- assert(model2.coefficients ~= coefficientsR2 relTol 1E-3)
+ assert(model1.intercept ~== interceptRStd relTol 1E-3)
+ assert(model1.coefficients ~= coefficientsRStd relTol 1E-3)
+ assert(model2.intercept ~== interceptR relTol 1E-3)
+ assert(model2.coefficients ~= coefficientsR relTol 1E-3)
}
test("binary logistic regression without intercept with L2 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(true)
+ .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(false)
+ .setElasticNetParam(0.0).setRegParam(1.37).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
+ Use the following R code to load the data and train the model using glmnet package.
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0, lambda = 1.37,
- intercept=FALSE))
- coefficients
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0,
+ lambda = 1.37, intercept=F, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0,
+ lambda = 1.37, intercept=F, standardize=F))
+ coefficientsStd
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) .
+ data.V3 -0.06000152
+ data.V4 0.12598737
+ data.V5 -0.04669009
+ data.V6 -0.09941025
- 5 x 1 sparse Matrix of class "dgCMatrix"
+ coefficients
+ 5 x 1 sparse Matrix of class "dgCMatrix"
s0
- (Intercept) .
- data.V2 -0.06099165
- data.V3 0.12857058
- data.V4 -0.04708770
- data.V5 -0.09799775
- */
- val interceptR1 = 0.0
- val coefficientsR1 = Vectors.dense(-0.06099165, 0.12857058, -0.04708770, -0.09799775)
-
- assert(model1.intercept ~== interceptR1 absTol 1E-3)
- assert(model1.coefficients ~= coefficientsR1 relTol 1E-2)
-
- /*
- Using the following R code to load the data and train the model using glmnet package.
-
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0, lambda = 1.37,
- intercept=FALSE, standardize=FALSE))
- coefficients
+ (Intercept) .
+ data.V3 -0.005482255
+ data.V4 0.048106338
+ data.V5 -0.093411640
+ data.V6 -0.054149798
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) .
- data.V2 -0.005679651
- data.V3 0.048967094
- data.V4 -0.093714016
- data.V5 -0.053314311
*/
- val interceptR2 = 0.0
- val coefficientsR2 = Vectors.dense(-0.005679651, 0.048967094, -0.093714016, -0.053314311)
+ val coefficientsRStd = Vectors.dense(-0.06000152, 0.12598737, -0.04669009, -0.09941025)
+ val coefficientsR = Vectors.dense(-0.005482255, 0.048106338, -0.093411640, -0.054149798)
- assert(model2.intercept ~== interceptR2 absTol 1E-3)
- assert(model2.coefficients ~= coefficientsR2 relTol 1E-2)
+ assert(model1.intercept ~== 0.0 absTol 1E-3)
+ assert(model1.coefficients ~= coefficientsRStd relTol 1E-2)
+ assert(model2.intercept ~== 0.0 absTol 1E-3)
+ assert(model2.coefficients ~= coefficientsR relTol 1E-2)
}
test("binary logistic regression with intercept with ElasticNet regularization") {
- val trainer1 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(true)
+ val trainer1 = (new LogisticRegression).setFitIntercept(true).setMaxIter(200)
+ .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(false)
+ .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
-
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0.38, lambda = 0.21))
- coefficients
-
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) 0.57734851
- data.V2 -0.05310287
- data.V3 .
- data.V4 -0.08849250
- data.V5 -0.15458796
- */
- val interceptR1 = 0.57734851
- val coefficientsR1 = Vectors.dense(-0.05310287, 0.0, -0.08849250, -0.15458796)
-
- assert(model1.intercept ~== interceptR1 relTol 6E-3)
- assert(model1.coefficients ~== coefficientsR1 absTol 5E-3)
+ Use the following R code to load the data and train the model using glmnet package.
- /*
- Using the following R code to load the data and train the model using glmnet package.
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0.38,
+ lambda = 0.21, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0.38,
+ lambda = 0.21, standardize=F))
+ coefficientsStd
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) 0.49991996
+ data.V3 -0.04131110
+ data.V4 .
+ data.V5 -0.08585233
+ data.V6 -0.15875400
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0.38, lambda = 0.21,
- standardize=FALSE))
- coefficients
+ coefficients
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) 0.5024256
+ data.V3 .
+ data.V4 .
+ data.V5 -0.1846038
+ data.V6 -0.0559614
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) 0.51555993
- data.V2 .
- data.V3 .
- data.V4 -0.18807395
- data.V5 -0.05350074
*/
- val interceptR2 = 0.51555993
- val coefficientsR2 = Vectors.dense(0.0, 0.0, -0.18807395, -0.05350074)
-
- assert(model2.intercept ~== interceptR2 relTol 6E-3)
- assert(model2.coefficients ~= coefficientsR2 absTol 1E-3)
+ val coefficientsRStd = Vectors.dense(-0.04131110, 0.0, -0.08585233, -0.15875400)
+ val interceptRStd = 0.49991996
+ val coefficientsR = Vectors.dense(0.0, 0.0, -0.1846038, -0.0559614)
+ val interceptR = 0.5024256
+
+ assert(model1.intercept ~== interceptRStd relTol 6E-3)
+ assert(model1.coefficients ~== coefficientsRStd absTol 5E-3)
+ assert(model2.intercept ~== interceptR relTol 6E-3)
+ assert(model2.coefficients ~= coefficientsR absTol 1E-3)
}
test("binary logistic regression without intercept with ElasticNet regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(true)
+ .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(false)
+ .setElasticNetParam(0.38).setRegParam(0.21).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
-
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0.38, lambda = 0.21,
- intercept=FALSE))
- coefficients
-
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) .
- data.V2 -0.001005743
- data.V3 0.072577857
- data.V4 -0.081203769
- data.V5 -0.142534158
- */
- val interceptR1 = 0.0
- val coefficientsR1 = Vectors.dense(-0.001005743, 0.072577857, -0.081203769, -0.142534158)
-
- assert(model1.intercept ~== interceptR1 relTol 1E-3)
- assert(model1.coefficients ~= coefficientsR1 absTol 1E-2)
+ Use the following R code to load the data and train the model using glmnet package.
- /*
- Using the following R code to load the data and train the model using glmnet package.
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0.38,
+ lambda = 0.21, intercept=FALSE, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 0.38,
+ lambda = 0.21, intercept=FALSE, standardize=F))
+ coefficientsStd
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) .
+ data.V3 .
+ data.V4 0.06859390
+ data.V5 -0.07900058
+ data.V6 -0.14684320
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 0.38, lambda = 0.21,
- intercept=FALSE, standardize=FALSE))
- coefficients
+ coefficients
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ (Intercept) .
+ data.V3 .
+ data.V4 0.03060637
+ data.V5 -0.11126742
+ data.V6 .
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) .
- data.V2 .
- data.V3 0.03345223
- data.V4 -0.11304532
- data.V5 .
*/
- val interceptR2 = 0.0
- val coefficientsR2 = Vectors.dense(0.0, 0.03345223, -0.11304532, 0.0)
+ val coefficientsRStd = Vectors.dense(0.0, 0.06859390, -0.07900058, -0.14684320)
+ val coefficientsR = Vectors.dense(0.0, 0.03060637, -0.11126742, 0.0)
- assert(model2.intercept ~== interceptR2 absTol 1E-3)
- assert(model2.coefficients ~= coefficientsR2 absTol 1E-3)
+ assert(model1.intercept ~== 0.0 relTol 1E-3)
+ assert(model1.coefficients ~= coefficientsRStd absTol 1E-2)
+ assert(model2.intercept ~== 0.0 absTol 1E-3)
+ assert(model2.coefficients ~= coefficientsR absTol 1E-3)
}
test("binary logistic regression with intercept with strong L1 regularization") {
- val trainer1 = (new LogisticRegression).setFitIntercept(true)
+ val trainer1 = (new LogisticRegression).setFitIntercept(true).setWeightCol("weight")
.setElasticNetParam(1.0).setRegParam(6.0).setStandardization(true)
- val trainer2 = (new LogisticRegression).setFitIntercept(true)
+ val trainer2 = (new LogisticRegression).setFitIntercept(true).setWeightCol("weight")
.setElasticNetParam(1.0).setRegParam(6.0).setStandardization(false)
val model1 = trainer1.fit(binaryDataset)
val model2 = trainer2.fit(binaryDataset)
- val histogram = binaryDataset.rdd.map { case Row(label: Double, features: Vector) => label }
+ val histogram = binaryDataset.as[Instance].rdd.map { i => (i.label, i.weight)}
.treeAggregate(new MultiClassSummarizer)(
seqOp = (c, v) => (c, v) match {
- case (classSummarizer: MultiClassSummarizer, label: Double) => classSummarizer.add(label)
+ case (classSummarizer: MultiClassSummarizer, (label: Double, weight: Double)) =>
+ classSummarizer.add(label, weight)
},
combOp = (c1, c2) => (c1, c2) match {
case (classSummarizer1: MultiClassSummarizer, classSummarizer2: MultiClassSummarizer) =>
@@ -989,25 +955,26 @@ class LogisticRegressionSuite
assert(model2.coefficients ~= coefficientsTheory absTol 1E-6)
/*
- TODO: why is this needed? The correctness of L1 regularization is already checked elsewhere
Using the following R code to load the data and train the model using glmnet package.
library("glmnet")
data <- read.csv("path", header=FALSE)
label = factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features,label, family="binomial", alpha = 1.0, lambda = 6.0))
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="binomial", alpha = 1.0,
+ lambda = 6.0))
coefficients
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- (Intercept) -0.2480643
- data.V2 0.0000000
- data.V3 .
- data.V4 .
- data.V5 .
+ s0
+ (Intercept) -0.2516986
+ data.V3 0.0000000
+ data.V4 .
+ data.V5 .
+ data.V6 .
*/
- val interceptR = -0.248065
+ val interceptR = -0.2516986
val coefficientsR = Vectors.dense(0.0, 0.0, 0.0, 0.0)
assert(model1.intercept ~== interceptR relTol 1E-5)
@@ -1015,9 +982,9 @@ class LogisticRegressionSuite
}
test("multinomial logistic regression with intercept with strong L1 regularization") {
- val trainer1 = (new LogisticRegression).setFitIntercept(true)
+ val trainer1 = (new LogisticRegression).setFitIntercept(true).setWeightCol("weight")
.setElasticNetParam(1.0).setRegParam(6.0).setStandardization(true)
- val trainer2 = (new LogisticRegression).setFitIntercept(true)
+ val trainer2 = (new LogisticRegression).setFitIntercept(true).setWeightCol("weight")
.setElasticNetParam(1.0).setRegParam(6.0).setStandardization(false)
val sqlContext = multinomialDataset.sqlContext
@@ -1025,16 +992,17 @@ class LogisticRegressionSuite
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
- val histogram = multinomialDataset.as[LabeledPoint].rdd.map(_.label)
+ val histogram = multinomialDataset.as[Instance].rdd.map(i => (i.label, i.weight))
.treeAggregate(new MultiClassSummarizer)(
seqOp = (c, v) => (c, v) match {
- case (classSummarizer: MultiClassSummarizer, label: Double) => classSummarizer.add(label)
+ case (classSummarizer: MultiClassSummarizer, (label: Double, weight: Double)) =>
+ classSummarizer.add(label, weight)
},
combOp = (c1, c2) => (c1, c2) match {
case (classSummarizer1: MultiClassSummarizer, classSummarizer2: MultiClassSummarizer) =>
classSummarizer1.merge(classSummarizer2)
}).histogram
- val numFeatures = multinomialDataset.as[LabeledPoint].first().features.size
+ val numFeatures = multinomialDataset.as[Instance].first().features.size
val numClasses = histogram.length
/*
@@ -1068,52 +1036,58 @@ class LogisticRegressionSuite
test("multinomial logistic regression with intercept without regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(true).setMaxIter(100)
+ .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(false)
+ .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
- > library("glmnet")
- > data <- read.csv("path", header=FALSE)
- > label = as.factor(data$V1)
- > features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- > coefficients = coef(glmnet(features, label, family="multinomial", alpha = 0, lambda = 0))
- > coefficients
- $`0`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -2.24493379
- V2 0.25096771
- V3 -0.03915938
- V4 0.14766639
- V5 0.36810817
- $`1`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.3778931
- V2 -0.3327489
- V3 0.8893666
- V4 -0.2306948
- V5 -0.4442330
- $`2`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 1.86704066
- V2 0.08178121
- V3 -0.85020722
- V4 0.08302840
- V5 0.07612480
- */
+ Use the following R code to load the data and train the model using glmnet package.
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = as.factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial",
+ alpha = 0, lambda = 0))
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ -2.10320093
+ data.V3 0.24337896
+ data.V4 -0.05916156
+ data.V5 0.14446790
+ data.V6 0.35976165
+
+ $`1`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ 0.3394473
+ data.V3 -0.3443375
+ data.V4 0.9181331
+ data.V5 -0.2283959
+ data.V6 -0.4388066
+
+ $`2`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ 1.76375361
+ data.V3 0.10095851
+ data.V4 -0.85897154
+ data.V5 0.08392798
+ data.V6 0.07904499
+
+
+ */
val coefficientsR = new DenseMatrix(3, 4, Array(
- 0.2509677, -0.0391594, 0.1476664, 0.3681082,
- -0.3327489, 0.8893666, -0.2306948, -0.4442330,
- 0.0817812, -0.8502072, 0.0830284, 0.0761248), isTransposed = true)
- val interceptsR = Vectors.dense(-2.2449338, 0.3778931, 1.8670407)
+ 0.24337896, -0.05916156, 0.14446790, 0.35976165,
+ -0.3443375, 0.9181331, -0.2283959, -0.4388066,
+ 0.10095851, -0.85897154, 0.08392798, 0.07904499), isTransposed = true)
+ val interceptsR = Vectors.dense(-2.10320093, 0.3394473, 1.76375361)
assert(model1.coefficientMatrix ~== coefficientsR relTol 0.05)
assert(model1.coefficientMatrix.toArray.sum ~== 0.0 absTol eps)
@@ -1128,52 +1102,57 @@ class LogisticRegressionSuite
test("multinomial logistic regression without intercept without regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(true)
+ .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(false)
+ .setElasticNetParam(0.0).setRegParam(0.0).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
/*
- Using the following R code to load the data and train the model using glmnet package.
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 0, lambda = 0,
- intercept=F))
- > coefficients
- $`0`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.06992464
- V3 -0.36562784
- V4 0.12142680
- V5 0.32052211
- $`1`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 -0.3036269
- V3 0.9449630
- V4 -0.2271038
- V5 -0.4364839
- $`2`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.2337022
- V3 -0.5793351
- V4 0.1056770
- V5 0.1159618
- */
+ Use the following R code to load the data and train the model using glmnet package.
+
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = as.factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0,
+ lambda = 0, intercept=F))
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ .
+ data.V3 0.07276291
+ data.V4 -0.36325496
+ data.V5 0.12015088
+ data.V6 0.31397340
+
+ $`1`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ .
+ data.V3 -0.3180040
+ data.V4 0.9679074
+ data.V5 -0.2252219
+ data.V6 -0.4319914
+ $`2`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ .
+ data.V3 0.2452411
+ data.V4 -0.6046524
+ data.V5 0.1050710
+ data.V6 0.1180180
+
+
+ */
val coefficientsR = new DenseMatrix(3, 4, Array(
- 0.0699246, -0.3656278, 0.1214268, 0.3205221,
- -0.3036269, 0.9449630, -0.2271038, -0.4364839,
- 0.2337022, -0.5793351, 0.1056770, 0.1159618), isTransposed = true)
+ 0.07276291, -0.36325496, 0.12015088, 0.31397340,
+ -0.3180040, 0.9679074, -0.2252219, -0.4319914,
+ 0.2452411, -0.6046524, 0.1050710, 0.1180180), isTransposed = true)
assert(model1.coefficientMatrix ~== coefficientsR relTol 0.05)
assert(model1.coefficientMatrix.toArray.sum ~== 0.0 absTol eps)
@@ -1190,92 +1169,95 @@ class LogisticRegressionSuite
// use tighter constraints because OWL-QN solver takes longer to converge
val trainer1 = (new LogisticRegression).setFitIntercept(true)
.setElasticNetParam(1.0).setRegParam(0.05).setStandardization(true)
- .setMaxIter(300).setTol(1e-10)
+ .setMaxIter(300).setTol(1e-10).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true)
.setElasticNetParam(1.0).setRegParam(0.05).setStandardization(false)
- .setMaxIter(300).setTol(1e-10)
+ .setMaxIter(300).setTol(1e-10).setWeightCol("weight")
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
/*
- Use the following R code to load the data and train the model using glmnet package.
- library("glmnet")
- data <- read.csv("path", header=FALSE)
- label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficientsStd = coef(glmnet(features, label, family="multinomial", alpha = 1,
- lambda = 0.05, standardization=T))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 1, lambda = 0.05,
- standardization=F))
- > coefficientsStd
- $`0`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.68988825
- V2 .
- V3 .
- V4 .
- V5 0.09404023
-
- $`1`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.2303499
- V2 -0.1232443
- V3 0.3258380
- V4 -0.1564688
- V5 -0.2053965
-
- $`2`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.9202381
- V2 .
- V3 -0.4803856
- V4 .
- V5 .
-
- > coefficients
- $`0`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.44893320
- V2 .
- V3 .
- V4 0.01933812
- V5 0.03666044
-
- $`1`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.7376760
- V2 -0.0577182
- V3 .
- V4 -0.2081718
- V5 -0.1304592
-
- $`2`
- 5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.2887428
- V2 .
- V3 .
- V4 .
- V5 .
- */
+ Use the following R code to load the data and train the model using glmnet package.
- val coefficientsRStd = new DenseMatrix(3, 4, Array(
- 0.0, 0.0, 0.0, 0.09404023,
- -0.1232443, 0.3258380, -0.1564688, -0.2053965,
- 0.0, -0.4803856, 0.0, 0.0), isTransposed = true)
- val interceptsRStd = Vectors.dense(-0.68988825, -0.2303499, 0.9202381)
+ library("glmnet")
+ data <- read.csv("path", header=FALSE)
+ label = as.factor(data$V1)
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="multinomial",
+ alpha = 1, lambda = 0.05, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 1,
+ lambda = 0.05, standardize=F))
+ coefficientsStd
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ -0.62244703
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 0.08419825
+ $`1`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ -0.2804845
+ data.V3 -0.1336960
+ data.V4 0.3717091
+ data.V5 -0.1530363
+ data.V6 -0.2035286
+
+ $`2`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ 0.9029315
+ data.V3 .
+ data.V4 -0.4629737
+ data.V5 .
+ data.V6 .
+
+
+ coefficients
+ $`0`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ -0.44215290
+ data.V3 .
+ data.V4 .
+ data.V5 0.01767089
+ data.V6 0.02542866
+
+ $`1`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ 0.76308326
+ data.V3 -0.06818576
+ data.V4 .
+ data.V5 -0.20446351
+ data.V6 -0.13017924
+
+ $`2`
+ 5 x 1 sparse Matrix of class "dgCMatrix"
+ s0
+ -0.3209304
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
+
+
+ */
+ val coefficientsRStd = new DenseMatrix(3, 4, Array(
+ 0.0, 0.0, 0.0, 0.08419825,
+ -0.1336960, 0.3717091, -0.1530363, -0.2035286,
+ 0.0, -0.4629737, 0.0, 0.0), isTransposed = true)
+ val interceptsRStd = Vectors.dense(-0.62244703, -0.2804845, 0.9029315)
val coefficientsR = new DenseMatrix(3, 4, Array(
- 0.0, 0.0, 0.01933812, 0.03666044,
- -0.0577182, 0.0, -0.2081718, -0.1304592,
+ 0.0, 0.0, 0.01767089, 0.02542866,
+ -0.06818576, 0.0, -0.20446351, -0.13017924,
0.0, 0.0, 0.0, 0.0), isTransposed = true)
- val interceptsR = Vectors.dense(-0.44893320, 0.7376760, -0.2887428)
+ val interceptsR = Vectors.dense(-0.44215290, 0.76308326, -0.3209304)
assert(model1.coefficientMatrix ~== coefficientsRStd absTol 0.02)
assert(model1.interceptVector ~== interceptsRStd relTol 0.1)
@@ -1287,87 +1269,91 @@ class LogisticRegressionSuite
test("multinomial logistic regression without intercept with L1 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(1.0).setRegParam(0.05).setStandardization(true)
+ .setElasticNetParam(1.0).setRegParam(0.05).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(1.0).setRegParam(0.05).setStandardization(false)
+ .setElasticNetParam(1.0).setRegParam(0.05).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
/*
Use the following R code to load the data and train the model using glmnet package.
+
library("glmnet")
data <- read.csv("path", header=FALSE)
label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficientsStd = coef(glmnet(features, label, family="multinomial", alpha = 1,
- lambda = 0.05, intercept=F, standardization=T))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 1, lambda = 0.05,
- intercept=F, standardization=F))
- > coefficientsStd
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 1,
+ lambda = 0.05, intercept=F, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 1,
+ lambda = 0.05, intercept=F, standardize=F))
+ coefficientsStd
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 .
- V4 .
- V5 0.01525105
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 0.01144225
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 -0.1502410
- V3 0.5134658
- V4 -0.1601146
- V5 -0.2500232
+ s0
+ .
+ data.V3 -0.1678787
+ data.V4 0.5385351
+ data.V5 -0.1573039
+ data.V6 -0.2471624
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.003301875
- V3 .
- V4 .
- V5 .
-
- > coefficients
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
+
+
+ coefficients
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 .
- V4 .
- V5 .
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 0.1943624
- V4 -0.1902577
- V5 -0.1028789
+ s0
+ .
+ data.V3 .
+ data.V4 0.1929409
+ data.V5 -0.1889121
+ data.V6 -0.1010413
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 .
- V4 .
- V5 .
- */
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
+
+ */
val coefficientsRStd = new DenseMatrix(3, 4, Array(
- 0.0, 0.0, 0.0, 0.01525105,
- -0.1502410, 0.5134658, -0.1601146, -0.2500232,
- 0.003301875, 0.0, 0.0, 0.0), isTransposed = true)
+ 0.0, 0.0, 0.0, 0.01144225,
+ -0.1678787, 0.5385351, -0.1573039, -0.2471624,
+ 0.0, 0.0, 0.0, 0.0), isTransposed = true)
val coefficientsR = new DenseMatrix(3, 4, Array(
0.0, 0.0, 0.0, 0.0,
- 0.0, 0.1943624, -0.1902577, -0.1028789,
+ 0.0, 0.1929409, -0.1889121, -0.1010413,
0.0, 0.0, 0.0, 0.0), isTransposed = true)
assert(model1.coefficientMatrix ~== coefficientsRStd absTol 0.01)
@@ -1380,92 +1366,95 @@ class LogisticRegressionSuite
test("multinomial logistic regression with intercept with L2 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(true)
+ .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(true)
- .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(false)
+ .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
/*
Use the following R code to load the data and train the model using glmnet package.
+
library("glmnet")
data <- read.csv("path", header=FALSE)
label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficientsStd = coef(glmnet(features, label, family="multinomial", alpha = 0,
- lambda = 0.1, intercept=T, standardization=T))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 0,
- lambda = 0.1, intercept=T, standardization=F))
- > coefficientsStd
+ w = data$V2
+ features = as.matrix(data.frame( data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="multinomial",
+ alpha = 0, lambda = 0.1, intercept=T, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0,
+ lambda = 0.1, intercept=T, standardize=F))
+ coefficientsStd
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -1.70040424
- V2 0.17576070
- V3 0.01527894
- V4 0.10216108
- V5 0.26099531
+ s0
+ -1.5898288335
+ data.V3 0.1691226336
+ data.V4 0.0002983651
+ data.V5 0.1001732896
+ data.V6 0.2554575585
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.2438590
- V2 -0.2238875
- V3 0.5967610
- V4 -0.1555496
- V5 -0.3010479
+ s0
+ 0.2125746
+ data.V3 -0.2304586
+ data.V4 0.6153492
+ data.V5 -0.1537017
+ data.V6 -0.2975443
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 1.45654525
- V2 0.04812679
- V3 -0.61203992
- V4 0.05338850
- V5 0.04005258
-
- > coefficients
+ s0
+ 1.37725427
+ data.V3 0.06133600
+ data.V4 -0.61564761
+ data.V5 0.05352840
+ data.V6 0.04208671
+
+
+ coefficients
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -1.65488543
- V2 0.15715048
- V3 0.01992903
- V4 0.12428858
- V5 0.22130317
+ s0
+ -1.5681088
+ data.V3 0.1508182
+ data.V4 0.0121955
+ data.V5 0.1217930
+ data.V6 0.2162850
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 1.1297533
- V2 -0.1974768
- V3 0.2776373
- V4 -0.1869445
- V5 -0.2510320
+ s0
+ 1.1217130
+ data.V3 -0.2028984
+ data.V4 0.2862431
+ data.V5 -0.1843559
+ data.V6 -0.2481218
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.52513212
- V2 0.04032627
- V3 -0.29756637
- V4 0.06265594
- V5 0.02972883
- */
+ s0
+ 0.44639579
+ data.V3 0.05208012
+ data.V4 -0.29843864
+ data.V5 0.06256289
+ data.V6 0.03183676
- val coefficientsRStd = new DenseMatrix(3, 4, Array(
- 0.17576070, 0.01527894, 0.10216108, 0.26099531,
- -0.2238875, 0.5967610, -0.1555496, -0.3010479,
- 0.04812679, -0.61203992, 0.05338850, 0.04005258), isTransposed = true)
- val interceptsRStd = Vectors.dense(-1.70040424, 0.2438590, 1.45654525)
+ */
+ val coefficientsRStd = new DenseMatrix(3, 4, Array(
+ 0.1691226336, 0.0002983651, 0.1001732896, 0.2554575585,
+ -0.2304586, 0.6153492, -0.1537017, -0.2975443,
+ 0.06133600, -0.61564761, 0.05352840, 0.04208671), isTransposed = true)
+ val interceptsRStd = Vectors.dense(-1.5898288335, 0.2125746, 1.37725427)
val coefficientsR = new DenseMatrix(3, 4, Array(
- 0.15715048, 0.01992903, 0.12428858, 0.22130317,
- -0.1974768, 0.2776373, -0.1869445, -0.2510320,
- 0.04032627, -0.29756637, 0.06265594, 0.02972883), isTransposed = true)
- val interceptsR = Vectors.dense(-1.65488543, 1.1297533, 0.52513212)
+ 0.1508182, 0.0121955, 0.1217930, 0.2162850,
+ -0.2028984, 0.2862431, -0.1843559, -0.2481218,
+ 0.05208012, -0.29843864, 0.06256289, 0.03183676), isTransposed = true)
+ val interceptsR = Vectors.dense(-1.5681088, 1.1217130, 0.44639579)
- assert(model1.coefficientMatrix ~== coefficientsRStd relTol 0.05)
+ assert(model1.coefficientMatrix ~== coefficientsRStd absTol 0.001)
assert(model1.interceptVector ~== interceptsRStd relTol 0.05)
assert(model1.interceptVector.toArray.sum ~== 0.0 absTol eps)
assert(model2.coefficientMatrix ~== coefficientsR relTol 0.05)
@@ -1475,86 +1464,92 @@ class LogisticRegressionSuite
test("multinomial logistic regression without intercept with L2 regularization") {
val trainer1 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(true)
+ .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(true).setWeightCol("weight")
val trainer2 = (new LogisticRegression).setFitIntercept(false)
- .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(false)
+ .setElasticNetParam(0.0).setRegParam(0.1).setStandardization(false).setWeightCol("weight")
val model1 = trainer1.fit(multinomialDataset)
val model2 = trainer2.fit(multinomialDataset)
/*
Use the following R code to load the data and train the model using glmnet package.
+
library("glmnet")
data <- read.csv("path", header=FALSE)
label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficientsStd = coef(glmnet(features, label, family="multinomial", alpha = 0,
- lambda = 0.1, intercept=F, standardization=T))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 0,
- lambda = 0.1, intercept=F, standardization=F))
- > coefficientsStd
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0,
+ lambda = 0.1, intercept=F, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0,
+ lambda = 0.1, intercept=F, standardize=F))
+ coefficientsStd
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.03904171
- V3 -0.23354322
- V4 0.08288096
- V5 0.22706393
+ s0
+ .
+ data.V3 0.04048126
+ data.V4 -0.23075758
+ data.V5 0.08228864
+ data.V6 0.22277648
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 -0.2061848
- V3 0.6341398
- V4 -0.1530059
- V5 -0.2958455
+ s0
+ .
+ data.V3 -0.2149745
+ data.V4 0.6478666
+ data.V5 -0.1515158
+ data.V6 -0.2930498
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.16714312
- V3 -0.40059658
- V4 0.07012496
- V5 0.06878158
- > coefficients
+ s0
+ .
+ data.V3 0.17449321
+ data.V4 -0.41710901
+ data.V5 0.06922716
+ data.V6 0.07027332
+
+
+ coefficients
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 -0.005704542
- V3 -0.144466409
- V4 0.092080736
- V5 0.182927657
+ s0
+ .
+ data.V3 -0.003949652
+ data.V4 -0.142982415
+ data.V5 0.091439598
+ data.V6 0.179286241
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 -0.08469036
- V3 0.38996748
- V4 -0.16468436
- V5 -0.22522976
+ s0
+ .
+ data.V3 -0.09071124
+ data.V4 0.39752531
+ data.V5 -0.16233832
+ data.V6 -0.22206059
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.09039490
- V3 -0.24550107
- V4 0.07260362
- V5 0.04230210
+ s0
+ .
+ data.V3 0.09466090
+ data.V4 -0.25454290
+ data.V5 0.07089872
+ data.V6 0.04277435
+
+
*/
val coefficientsRStd = new DenseMatrix(3, 4, Array(
- 0.03904171, -0.23354322, 0.08288096, 0.2270639,
- -0.2061848, 0.6341398, -0.1530059, -0.2958455,
- 0.16714312, -0.40059658, 0.07012496, 0.06878158), isTransposed = true)
+ 0.04048126, -0.23075758, 0.08228864, 0.22277648,
+ -0.2149745, 0.6478666, -0.1515158, -0.2930498,
+ 0.17449321, -0.41710901, 0.06922716, 0.07027332), isTransposed = true)
val coefficientsR = new DenseMatrix(3, 4, Array(
- -0.005704542, -0.144466409, 0.092080736, 0.182927657,
- -0.08469036, 0.38996748, -0.16468436, -0.22522976,
- 0.0903949, -0.24550107, 0.07260362, 0.0423021), isTransposed = true)
+ -0.003949652, -0.142982415, 0.091439598, 0.179286241,
+ -0.09071124, 0.39752531, -0.16233832, -0.22206059,
+ 0.09466090, -0.25454290, 0.07089872, 0.04277435), isTransposed = true)
assert(model1.coefficientMatrix ~== coefficientsRStd absTol 0.01)
assert(model1.interceptVector.toArray === Array.fill(3)(0.0))
@@ -1565,10 +1560,10 @@ class LogisticRegressionSuite
}
test("multinomial logistic regression with intercept with elasticnet regularization") {
- val trainer1 = (new LogisticRegression).setFitIntercept(true)
+ val trainer1 = (new LogisticRegression).setFitIntercept(true).setWeightCol("weight")
.setElasticNetParam(0.5).setRegParam(0.1).setStandardization(true)
.setMaxIter(300).setTol(1e-10)
- val trainer2 = (new LogisticRegression).setFitIntercept(true)
+ val trainer2 = (new LogisticRegression).setFitIntercept(true).setWeightCol("weight")
.setElasticNetParam(0.5).setRegParam(0.1).setStandardization(false)
.setMaxIter(300).setTol(1e-10)
@@ -1576,82 +1571,85 @@ class LogisticRegressionSuite
val model2 = trainer2.fit(multinomialDataset)
/*
Use the following R code to load the data and train the model using glmnet package.
+
library("glmnet")
data <- read.csv("path", header=FALSE)
label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficientsStd = coef(glmnet(features, label, family="multinomial", alpha = 0.5,
- lambda = 0.1, intercept=T, standardization=T))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 0.5,
- lambda = 0.1, intercept=T, standardization=F))
- > coefficientsStd
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0.5,
+ lambda = 0.1, intercept=T, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0.5,
+ lambda = 0.1, intercept=T, standardize=F))
+ coefficientsStd
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.5521819483
- V2 0.0003092611
- V3 .
- V4 .
- V5 0.0913818490
+ s0
+ -0.50133383
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 0.08351653
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.27531989
- V2 -0.09790029
- V3 0.28502034
- V4 -0.12416487
- V5 -0.16513373
+ s0
+ -0.3151913
+ data.V3 -0.1058702
+ data.V4 0.3183251
+ data.V5 -0.1212969
+ data.V6 -0.1629778
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.8275018
- V2 .
- V3 -0.4044859
- V4 .
- V5 .
-
- > coefficients
+ s0
+ 0.8165252
+ data.V3 .
+ data.V4 -0.3943069
+ data.V5 .
+ data.V6 .
+
+
+ coefficients
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.39876213
- V2 .
- V3 .
- V4 0.02547520
- V5 0.03893991
+ s0
+ -0.38857157
+ data.V3 .
+ data.V4 .
+ data.V5 0.02384198
+ data.V6 0.03127749
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- 0.61089869
- V2 -0.04224269
- V3 .
- V4 -0.18923970
- V5 -0.09104249
+ s0
+ 0.62492165
+ data.V3 -0.04949061
+ data.V4 .
+ data.V5 -0.18584462
+ data.V6 -0.08952455
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- -0.2121366
- V2 .
- V3 .
- V4 .
- V5 .
- */
+ s0
+ -0.2363501
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
- val coefficientsRStd = new DenseMatrix(3, 4, Array(
- 0.0003092611, 0.0, 0.0, 0.091381849,
- -0.09790029, 0.28502034, -0.12416487, -0.16513373,
- 0.0, -0.4044859, 0.0, 0.0), isTransposed = true)
- val interceptsRStd = Vectors.dense(-0.5521819483, -0.27531989, 0.8275018)
+ */
+ val coefficientsRStd = new DenseMatrix(3, 4, Array(
+ 0.0, 0.0, 0.0, 0.08351653,
+ -0.1058702, 0.3183251, -0.1212969, -0.1629778,
+ 0.0, -0.3943069, 0.0, 0.0), isTransposed = true)
+ val interceptsRStd = Vectors.dense(-0.50133383, -0.3151913, 0.8165252)
val coefficientsR = new DenseMatrix(3, 4, Array(
- 0.0, 0.0, 0.0254752, 0.03893991,
- -0.04224269, 0.0, -0.1892397, -0.09104249,
+ 0.0, 0.0, 0.02384198, 0.03127749,
+ -0.04949061, 0.0, -0.18584462, -0.08952455,
0.0, 0.0, 0.0, 0.0), isTransposed = true)
- val interceptsR = Vectors.dense(-0.39876213, 0.61089869, -0.2121366)
+ val interceptsR = Vectors.dense(-0.38857157, 0.62492165, -0.2363501)
assert(model1.coefficientMatrix ~== coefficientsRStd absTol 0.01)
assert(model1.interceptVector ~== interceptsRStd absTol 0.01)
@@ -1662,10 +1660,10 @@ class LogisticRegressionSuite
}
test("multinomial logistic regression without intercept with elasticnet regularization") {
- val trainer1 = (new LogisticRegression).setFitIntercept(false)
+ val trainer1 = (new LogisticRegression).setFitIntercept(false).setWeightCol("weight")
.setElasticNetParam(0.5).setRegParam(0.1).setStandardization(true)
.setMaxIter(300).setTol(1e-10)
- val trainer2 = (new LogisticRegression).setFitIntercept(false)
+ val trainer2 = (new LogisticRegression).setFitIntercept(false).setWeightCol("weight")
.setElasticNetParam(0.5).setRegParam(0.1).setStandardization(false)
.setMaxIter(300).setTol(1e-10)
@@ -1673,78 +1671,83 @@ class LogisticRegressionSuite
val model2 = trainer2.fit(multinomialDataset)
/*
Use the following R code to load the data and train the model using glmnet package.
+
library("glmnet")
data <- read.csv("path", header=FALSE)
label = as.factor(data$V1)
- features = as.matrix(data.frame(data$V2, data$V3, data$V4, data$V5))
- coefficientsStd = coef(glmnet(features, label, family="multinomial", alpha = 0.5,
- lambda = 0.1, intercept=F, standardization=T))
- coefficients = coef(glmnet(features, label, family="multinomial", alpha = 0.5,
- lambda = 0.1, intercept=F, standardization=F))
- > coefficientsStd
+ w = data$V2
+ features = as.matrix(data.frame(data$V3, data$V4, data$V5, data$V6))
+ coefficientsStd = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0.5,
+ lambda = 0.1, intercept=F, standardize=T))
+ coefficients = coef(glmnet(features, label, weights=w, family="multinomial", alpha = 0.5,
+ lambda = 0.1, intercept=F, standardize=F))
+ coefficientsStd
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 .
- V4 .
- V5 0.03543706
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 0.03238285
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 -0.1187387
- V3 0.4025482
- V4 -0.1270969
- V5 -0.1918386
+ s0
+ .
+ data.V3 -0.1328284
+ data.V4 0.4219321
+ data.V5 -0.1247544
+ data.V6 -0.1893318
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 0.00774365
- V3 .
- V4 .
- V5 .
-
- > coefficients
+ s0
+ .
+ data.V3 0.004572312
+ data.V4 .
+ data.V5 .
+ data.V6 .
+
+
+ coefficients
$`0`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 .
- V4 .
- V5 .
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
$`1`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 0.14666497
- V4 -0.16570638
- V5 -0.05982875
+ s0
+ .
+ data.V3 .
+ data.V4 0.14571623
+ data.V5 -0.16456351
+ data.V6 -0.05866264
$`2`
5 x 1 sparse Matrix of class "dgCMatrix"
- s0
- .
- V2 .
- V3 .
- V4 .
- V5 .
+ s0
+ .
+ data.V3 .
+ data.V4 .
+ data.V5 .
+ data.V6 .
+
+
*/
val coefficientsRStd = new DenseMatrix(3, 4, Array(
- 0.0, 0.0, 0.0, 0.03543706,
- -0.1187387, 0.4025482, -0.1270969, -0.1918386,
- 0.0, 0.0, 0.0, 0.00774365), isTransposed = true)
+ 0.0, 0.0, 0.0, 0.03238285,
+ -0.1328284, 0.4219321, -0.1247544, -0.1893318,
+ 0.004572312, 0.0, 0.0, 0.0), isTransposed = true)
val coefficientsR = new DenseMatrix(3, 4, Array(
0.0, 0.0, 0.0, 0.0,
- 0.0, 0.14666497, -0.16570638, -0.05982875,
+ 0.0, 0.14571623, -0.16456351, -0.05866264,
0.0, 0.0, 0.0, 0.0), isTransposed = true)
assert(model1.coefficientMatrix ~== coefficientsRStd absTol 0.01)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@spark.apache.org
For additional commands, e-mail: commits-help@spark.apache.org