[spark] branch branch-3.1 updated: [MINOR][ML] Increase Bounded MLOR (without regularization) test error tolerance

[we...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

weichenxu123 pushed a commit to branch branch-3.1
in repository https://gitbox.apache.org/repos/asf/spark.git


The following commit(s) were added to refs/heads/branch-3.1 by this push:
     new b0a70ab  [MINOR][ML] Increase Bounded MLOR (without regularization) test error tolerance
b0a70ab is described below

commit b0a70abed383cedd59f28cec75aa898df0c0b4bd
Author: Weichen Xu <we...@databricks.com>
AuthorDate: Wed Dec 9 11:18:09 2020 +0800

    [MINOR][ML] Increase Bounded MLOR (without regularization) test error tolerance
    
    ### What changes were proposed in this pull request?
    Improve LogisticRegression test error tolerance
    
    ### Why are the changes needed?
    When we switch BLAS version, some of the tests will fail due to too strict error tolerance in test.
    
    ### Does this PR introduce _any_ user-facing change?
    No
    
    ### How was this patch tested?
    N/A
    
    Closes #30587 from WeichenXu123/fix_lor_test.
    
    Authored-by: Weichen Xu <we...@databricks.com>
    Signed-off-by: Weichen Xu <we...@databricks.com>
    (cherry picked from commit f021f6d3c72e1c84637798b4ddcb7e208fdfbf46)
    Signed-off-by: Weichen Xu <we...@databricks.com>
---
 .../ml/classification/LogisticRegressionSuite.scala | 21 +++++++++++----------
 1 file changed, 11 insertions(+), 10 deletions(-)

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 d0b282d..d2814b4 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
@@ -1548,9 +1548,9 @@ class LogisticRegressionSuite extends MLTest with DefaultReadWriteTest {
     val interceptsExpected1 = Vectors.dense(
       1.0000152482448372, 3.591773288423673, 5.079685953744937)
 
-    checkCoefficientsEquivalent(model1.coefficientMatrix, coefficientsExpected1)
+    checkBoundedMLORCoefficientsEquivalent(model1.coefficientMatrix, coefficientsExpected1)
     assert(model1.interceptVector ~== interceptsExpected1 relTol 0.01)
-    checkCoefficientsEquivalent(model2.coefficientMatrix, coefficientsExpected1)
+    checkBoundedMLORCoefficientsEquivalent(model2.coefficientMatrix, coefficientsExpected1)
     assert(model2.interceptVector ~== interceptsExpected1 relTol 0.01)
 
     // Bound constrained optimization with bound on both side.
@@ -1585,9 +1585,9 @@ class LogisticRegressionSuite extends MLTest with DefaultReadWriteTest {
       isTransposed = true)
     val interceptsExpected3 = Vectors.dense(1.0, 2.0, 2.0)
 
-    checkCoefficientsEquivalent(model3.coefficientMatrix, coefficientsExpected3)
+    checkBoundedMLORCoefficientsEquivalent(model3.coefficientMatrix, coefficientsExpected3)
     assert(model3.interceptVector ~== interceptsExpected3 relTol 0.01)
-    checkCoefficientsEquivalent(model4.coefficientMatrix, coefficientsExpected3)
+    checkBoundedMLORCoefficientsEquivalent(model4.coefficientMatrix, coefficientsExpected3)
     assert(model4.interceptVector ~== interceptsExpected3 relTol 0.01)
 
     // Bound constrained optimization with infinite bound on both side.
@@ -1621,9 +1621,9 @@ class LogisticRegressionSuite extends MLTest with DefaultReadWriteTest {
     val interceptsExpected5 = Vectors.dense(
       -2.2231282183460723, 0.3669496747012527, 1.856178543644802)
 
-    checkCoefficientsEquivalent(model5.coefficientMatrix, coefficientsExpected5)
+    checkBoundedMLORCoefficientsEquivalent(model5.coefficientMatrix, coefficientsExpected5)
     assert(model5.interceptVector ~== interceptsExpected5 relTol 0.01)
-    checkCoefficientsEquivalent(model6.coefficientMatrix, coefficientsExpected5)
+    checkBoundedMLORCoefficientsEquivalent(model6.coefficientMatrix, coefficientsExpected5)
     assert(model6.interceptVector ~== interceptsExpected5 relTol 0.01)
   }
 
@@ -1719,9 +1719,9 @@ class LogisticRegressionSuite extends MLTest with DefaultReadWriteTest {
       1.7524631428961193, 1.2292565990448736, 1.3433784431904323, 1.5846063017678864),
       isTransposed = true)
 
-    checkCoefficientsEquivalent(model1.coefficientMatrix, coefficientsExpected)
+    checkBoundedMLORCoefficientsEquivalent(model1.coefficientMatrix, coefficientsExpected)
     assert(model1.interceptVector.toArray === Array.fill(3)(0.0))
-    checkCoefficientsEquivalent(model2.coefficientMatrix, coefficientsExpected)
+    checkBoundedMLORCoefficientsEquivalent(model2.coefficientMatrix, coefficientsExpected)
     assert(model2.interceptVector.toArray === Array.fill(3)(0.0))
   }
 
@@ -2953,16 +2953,17 @@ object LogisticRegressionSuite {
   }
 
   /**
+   * Note: This method is only used in Bounded MLOR (without regularization) test
    * When no regularization is applied, the multinomial coefficients lack identifiability
    * because we do not use a pivot class. We can add any constant value to the coefficients
    * and get the same likelihood. If fitting under bound constrained optimization, we don't
    * choose the mean centered coefficients like what we do for unbound problems, since they
    * may out of the bounds. We use this function to check whether two coefficients are equivalent.
    */
-  def checkCoefficientsEquivalent(coefficients1: Matrix, coefficients2: Matrix): Unit = {
+  def checkBoundedMLORCoefficientsEquivalent(coefficients1: Matrix, coefficients2: Matrix): Unit = {
     coefficients1.colIter.zip(coefficients2.colIter).foreach { case (col1: Vector, col2: Vector) =>
       (col1.asBreeze - col2.asBreeze).toArray.toSeq.sliding(2).foreach {
-        case Seq(v1, v2) => assert(v1 ~= v2 absTol 1E-3)
+        case Seq(v1, v2) => assert(v1 ~= v2 absTol 1E-2)
       }
     }
   }


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@spark.apache.org
For additional commands, e-mail: commits-help@spark.apache.org