Github user yanboliang commented on a diff in the pull request: https://github.com/apache/spark/pull/17746#discussion_r113133446 --- Diff: R/pkg/inst/tests/testthat/test_mllib_classification.R --- @@ -288,18 +288,18 @@ test_that("spark.mlp", { c(0, 0, 0, 0, 0, 5, 5, 5, 5, 5, 9, 9, 9, 9, 9)) mlpPredictions <- collect(select(predict(model, mlpTestDF), "prediction")) expect_equal(head(mlpPredictions$prediction, 10), - c("1.0", "1.0", "1.0", "1.0", "2.0", "1.0", "2.0", "2.0", "1.0", "0.0")) + c("1.0", "1.0", "2.0", "1.0", "2.0", "1.0", "2.0", "2.0", "1.0", "0.0")) model <- spark.mlp(df, label ~ features, layers = c(4, 3), maxIter = 2, initialWeights = c(0.0, 0.0, 0.0, 0.0, 0.0, 5.0, 5.0, 5.0, 5.0, 5.0, 9.0, 9.0, 9.0, 9.0, 9.0)) mlpPredictions <- collect(select(predict(model, mlpTestDF), "prediction")) expect_equal(head(mlpPredictions$prediction, 10), - c("1.0", "1.0", "1.0", "1.0", "2.0", "1.0", "2.0", "2.0", "1.0", "0.0")) + c("1.0", "1.0", "2.0", "1.0", "2.0", "1.0", "2.0", "2.0", "1.0", "0.0")) model <- spark.mlp(df, label ~ features, layers = c(4, 3), maxIter = 2) mlpPredictions <- collect(select(predict(model, mlpTestDF), "prediction")) expect_equal(head(mlpPredictions$prediction, 10), - c("1.0", "1.0", "1.0", "1.0", "0.0", "1.0", "0.0", "2.0", "1.0", "0.0")) + c("1.0", "1.0", "1.0", "1.0", "0.0", "1.0", "0.0", "0.0", "1.0", "0.0")) --- End diff -- Yeah, It's weird. I suspect that numerical issue or some underlying breeze fixes caused the output changed, the affected results only happened in very tiny dataset(three rows) in PySpark/SparkR, there is no effect for all Scala tests(usually thousands of rows). I run PySpark ```LogisticRegression``` on a larger dataset against Spark depends on breeze 0.12 and 0.13.1, they got the same result with reasonable tolerance: For breeze 0.12: ``` >>> df = spark.read.format("libsvm").load("/Users/yliang/data/trunk4/spark/data/mllib/sample_multiclass_classification_data.txt") >>> from pyspark.ml.classification import LogisticRegression >>> mlor = LogisticRegression(maxIter=100, regParam=0.01, family="multinomial") >>> mlorModel = mlor.fit(df) >>> mlorModel.coefficientMatrix DenseMatrix(3, 4, [1.0584, -1.8365, 3.2426, 3.6224, -2.1275, 2.8712, -2.8362, -2.5096, 1.069, -1.0347, -0.4064, -1.1128], 1) >>> mlorModel.interceptVector DenseVector([-1.1036, -0.5917, 1.6953]) ``` For breeze 0.13.1: ``` >>> df = spark.read.format("libsvm").load("/Users/yliang/data/trunk4/spark/data/mllib/sample_multiclass_classification_data.txt") >>> from pyspark.ml.classification import LogisticRegression >>> mlor = LogisticRegression(maxIter=100, regParam=0.01, family="multinomial") >>> mlorModel = mlor.fit(df) >>> mlorModel.coefficientMatrix DenseMatrix(3, 4, [1.0584, -1.8365, 3.2426, 3.6224, -2.1274, 2.8712, -2.8363, -2.5096, 1.069, -1.0347, -0.4064, -1.1128], 1) >>> mlorModel.interceptVector DenseVector([-1.1036, -0.5917, 1.6953]) ``` So I think we should update these vulnerable PySpark/SparkR test cases to use larger dataset to make them more stable. What about merging this firstly and do that in a separate PR? Thanks.
--- 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