[SPARK-17241] add sanity check

Author: keypointt

R/pkg/inst/tests/testthat/test_mllib.R

@@ -99,10 +99,6 @@ test_that("spark.glm summary", {
   expect_match(out[2], "Deviance Residuals:")
   expect_true(any(grepl("AIC: 59.22", out)))
 
-  # Test spark.glm works with regularization parameter
-  regStats <- summary(spark.glm(training, Sepal_Width ~ Sepal_Length + Species, regParam = 0.3))
-  expect_equal(regStats$aic, 136.7, tolerance = 1e-3)
-
   # binomial family
   df <- suppressWarnings(createDataFrame(iris))
   training <- df[df$Species %in% c("versicolor", "virginica"), ]
@@ -152,6 +148,12 @@ test_that("spark.glm summary", {
   baseModel <- stats::glm(Sepal.Width ~ Sepal.Length + Species, data = iris)
   baseSummary <- summary(baseModel)
   expect_true(abs(baseSummary$deviance - 12.19313) < 1e-4)
+
+  # Test spark.glm works with regularization parameter
+  data <- as.data.frame(cbind(a1, a2, b))
+  df <- suppressWarnings(createDataFrame(data))
+  regStats <- summary(spark.glm(df, b ~ a1 + a2, regParam = 1.0))
+  expect_equal(regStats$aic, 13.32836, tolerance = 1e-4) # 13.32836 is from summary() result
 })
 
 test_that("spark.glm save/load", {

mllib/src/test/scala/org/apache/spark/ml/regression/GeneralizedLinearRegressionSuite.scala

@@ -1034,6 +1034,46 @@ class GeneralizedLinearRegressionSuite
       .setFamily("gaussian")
       .fit(datasetGaussianIdentity.as[LabeledPoint])
   }
+
+  test("generalized linear regression: regularization parameter") {
+    /*
+      R code:
+
+      a1 <- c(0, 1, 2, 3)
+      a2 <- c(5, 2, 1, 3)
+      b <- c(1, 0, 1, 0)
+      data <- as.data.frame(cbind(a1, a2, b))
+      df <- suppressWarnings(createDataFrame(data))
+
+      for (regParam in c(0.0, 0.1, 1.0)) {
+        model <- spark.glm(df, b ~ a1 + a2, regParam = regParam)
+        print(as.vector(summary(model)$aic))
+      }
+
+      [1] 12.88188
+      [1] 12.92681
+      [1] 13.32836
+     */
+    val dataset = spark.createDataFrame(Seq(
+      LabeledPoint(1, Vectors.dense(5, 0)),
+      LabeledPoint(0, Vectors.dense(2, 1)),
+      LabeledPoint(1, Vectors.dense(1, 2)),
+      LabeledPoint(0, Vectors.dense(3, 3))
+    ))
+    val expected = Seq(12.88188, 12.92681, 13.32836)
+
+    var idx = 0
+    for (regParam <- Seq(0.0, 0.1, 1.0)) {
+      val trainer = new GeneralizedLinearRegression()
+        .setRegParam(regParam)
+        .setLabelCol("label")
+        .setFeaturesCol("features")
+      val model = trainer.fit(dataset)
+      val actual = model.summary.aic
+      assert(actual ~= expected(idx) absTol 1e-4, "Model mismatch: GLM with regParam = $regParam.")
+      idx += 1
+    }
+  }
 }
 
 object GeneralizedLinearRegressionSuite {