Github user sethah commented on a diff in the pull request:

    https://github.com/apache/spark/pull/15394#discussion_r83081588
  
    --- Diff: 
mllib/src/test/scala/org/apache/spark/ml/optim/WeightedLeastSquaresSuite.scala 
---
    @@ -60,24 +60,87 @@ class WeightedLeastSquaresSuite extends SparkFunSuite 
with MLlibTestSparkContext
         ), 2)
       }
     
    -  test("two collinear features result in error with no regularization") {
    +  test("WLS with strong L1 regularization") {
    +    /*
    +      We initialize the coefficients for WLS QN solver to be weighted 
average of the label. Check
    +      here that with only an intercept the model converges to bBar.
    +     */
    +    val bAgg = instances.collect().foldLeft((0.0, 0.0)) { case ((sum, 
count), Instance(l, w, f)) =>
    +      (sum + w * l, count + w)
    +    }
    +    val bBar = bAgg._1 / bAgg._2
    +    val wls = new WeightedLeastSquares(true, 10, 1.0, true, true)
    +    val model = wls.fit(instances)
    +    assert(model.intercept ~== bBar relTol 1e-6)
    +  }
    +
    +  test("diagonal inverse of AtWA") {
    +    /*
    +      A <- matrix(c(0, 1, 2, 3, 5, 7, 11, 13), 4, 2)
    +      w <- c(1, 2, 3, 4)
    +      W <- Diagonal(length(w), w)
    +      A.intercept <- cbind(A, rep.int(1, length(w)))
    +      AtA.intercept <- t(A.intercept) %*% W %*% A.intercept
    +      inv.intercept <- solve(AtA.intercept)
    +      print(diag(inv.intercept))
    +      [1]  4.02  0.50 12.02
    +
    +      AtA <- t(A) %*% W %*% A
    +      inv <- solve(AtA)
    +      print(diag(inv))
    +      [1] 0.48336106 0.02079867
    +
    +     */
    +    val expectedWithIntercept = Vectors.dense(4.02, 0.50, 12.02)
    +    val expected = Vectors.dense(0.48336106, 0.02079867)
    +    val wlsWithIntercept = new WeightedLeastSquares(true, 0.0, 0.0, true, 
true,
    --- End diff --
    
    Yeah, I thought about this. It makes the code kind of ridiculously long. 
Done.


---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org

Reply via email to