BatchNormal anomaly detector throws "java.lang.IncompatibleClassChangeError: Implementing class"

[Blackbaud-PatrickMcDonald]

Environment: Databricks Spark 3.0.1 Scala 2.12.10 Java 1.8.0_282 deequ:1.2.2-spark-3.0 pydeequ 1.0.1

I'm attempting to use the BatchNormal anomaly detector strategy and it is throwing the following exception:

Py4JJavaError: An error occurred while calling o405.run.
: java.lang.IncompatibleClassChangeError: Implementing class
    at com.amazon.deequ.anomalydetection.BatchNormalStrategy.detect(BatchNormalStrategy.scala:75)
    at com.amazon.deequ.anomalydetection.AnomalyDetector.detectAnomaliesInHistory(AnomalyDetector.scala:98)
    at com.amazon.deequ.anomalydetection.AnomalyDetector.isNewPointAnomalous(AnomalyDetector.scala:60)
    at com.amazon.deequ.checks.Check$.isNewestPointNonAnomalous(Check.scala:1138)
    at com.amazon.deequ.checks.Check.$anonfun$isNewestPointNonAnomalous$1(Check.scala:433)
    at scala.runtime.java8.JFunction1$mcZD$sp.apply(JFunction1$mcZD$sp.java:23)
    at com.amazon.deequ.constraints.AnalysisBasedConstraint.runAssertion(AnalysisBasedConstraint.scala:108)
    at com.amazon.deequ.constraints.AnalysisBasedConstraint.pickValueAndAssert(AnalysisBasedConstraint.scala:74)
    at com.amazon.deequ.constraints.AnalysisBasedConstraint.$anonfun$evaluate$2(AnalysisBasedConstraint.scala:60)
    at scala.Option.map(Option.scala:230)
    at com.amazon.deequ.constraints.AnalysisBasedConstraint.evaluate(AnalysisBasedConstraint.scala:60)
    at com.amazon.deequ.constraints.ConstraintDecorator.evaluate(Constraint.scala:56)
    at com.amazon.deequ.checks.Check.$anonfun$evaluate$1(Check.scala:1038)
    at scala.collection.TraversableLike.$anonfun$map$1(TraversableLike.scala:238)
    at scala.collection.immutable.List.foreach(List.scala:392)
    at scala.collection.TraversableLike.map(TraversableLike.scala:238)
    at scala.collection.TraversableLike.map$(TraversableLike.scala:231)
    at scala.collection.immutable.List.map(List.scala:298)
    at com.amazon.deequ.checks.Check.evaluate(Check.scala:1038)
    at com.amazon.deequ.VerificationSuite.$anonfun$evaluate$1(VerificationSuite.scala:269)
    at scala.collection.TraversableLike.$anonfun$map$1(TraversableLike.scala:238)
    at scala.collection.immutable.List.foreach(List.scala:392)
    at scala.collection.TraversableLike.map(TraversableLike.scala:238)
    at scala.collection.TraversableLike.map$(TraversableLike.scala:231)
    at scala.collection.immutable.List.map(List.scala:298)
    at com.amazon.deequ.VerificationSuite.evaluate(VerificationSuite.scala:269)
    at com.amazon.deequ.VerificationSuite.doVerificationRun(VerificationSuite.scala:132)
    at com.amazon.deequ.VerificationRunBuilder.run(VerificationRunBuilder.scala:173)
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:498)
    at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:244)
    at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:380)
    at py4j.Gateway.invoke(Gateway.java:295)
    at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
    at py4j.commands.CallCommand.execute(CallCommand.java:79)
    at py4j.GatewayConnection.run(GatewayConnection.java:251)
    at java.lang.Thread.run(Thread.java:748)

To reproduce:

from pyspark.sql import SparkSession, Row
from pydeequ.repository import *
from pydeequ.verification import *
from pydeequ.anomaly_detection import *
from pydeequ.analyzers import *

spark = (SparkSession
    .builder
    .getOrCreate())
sc = spark.sparkContext

metricsRepositoryPath = "dbfs:/path/to/my/metrics.json"
metricsRepository = FileSystemMetricsRepository(spark, metricsRepositoryPath)

todaysDataset = sc.parallelize([
            Row(a=80,b=5,),
            Row(a=3,  b=30,),
            Row(a=10, b=5,)]).toDF()

currTime = ResultKey.current_milli_time()
todaysKey = ResultKey(spark, currTime)

currResult = VerificationSuite(spark).onData(todaysDataset) \
    .useRepository(metricsRepository) \
    .addAnomalyCheck(BatchNormalStrategy(), Sum("a")) \
    .run()

Is this a compatibility issue with the version of Breeze being used?

[sakshaat]

Hi, I'm seeing a similar issue with this anomaly check as well on EMR 6.40 (spark 3.1.2) running deequ 2.0.0. This is the exception I get:

java.lang.NoClassDefFoundError: breeze/stats/package$
  at com.amazon.deequ.anomalydetection.BatchNormalStrategy.detect(BatchNormalStrategy.scala:75)
  at com.amazon.deequ.anomalydetection.AnomalyDetector.detectAnomaliesInHistory(AnomalyDetector.scala:98)
  at com.amazon.deequ.anomalydetection.AnomalyDetector.isNewPointAnomalous(AnomalyDetector.scala:60)
  at com.amazon.deequ.checks.Check$.isNewestPointNonAnomalous(Check.scala:1138)
  at com.amazon.deequ.checks.Check.$anonfun$isNewestPointNonAnomalous$1(Check.scala:433)
  at scala.runtime.java8.JFunction1$mcZD$sp.apply(JFunction1$mcZD$sp.java:23)
  at com.amazon.deequ.constraints.AnalysisBasedConstraint.runAssertion(AnalysisBasedConstraint.scala:108)
  at com.amazon.deequ.constraints.AnalysisBasedConstraint.pickValueAndAssert(AnalysisBasedConstraint.scala:74)
  at com.amazon.deequ.constraints.AnalysisBasedConstraint.$anonfun$evaluate$2(AnalysisBasedConstraint.scala:60)

import com.amazon.deequ.anomalydetection.BatchNormalStrategy

def batchNormalStrategyTest() = {
    val strategy = BatchNormalStrategy(lowerDeviationFactor=Some(3.0), upperDeviationFactor=Some(3.0))
    datasetVerificationRunner(df, {v: VerificationRunBuilderWithRepository => v.addAnomalyCheck(strategy, Compliance("column_1", "cardinality(column_1) > 0"))})
}

batchNormalStrategyTest()

The datasetVerificationRunner function is simply building a VerificationRunBuilderWithRepository and applying the higher order function on it to add an AnomalyCheck (It is structured this way because my intention is to quickly test out different strategies on a number of different instances of the same dataset).

Basically something like

val verificationResult = addAnomalyCheck(VerificationSuite().onData(df.where(s"date='${date}'")).useRepository(repo).saveOrAppendResult(newKey)).run()
val resultDf = checkResultsAsDataFrame(spark, verificationResult)

Is this a known issue with this class or am I using it incorrectly in this case?

[mfridrikhson-tp]

For anyone having trouble running it on Databricks - it seems to be a conflict between breeze dependency versions - DBR 7.3 and later use 1.x which conflicts with Deequ's 0.13.2

[chenliu0831]

Possible to verify if the issue go away with latest mainline?

pip install git+https://github.com/awslabs/python-deequ.git@7ec9f6f72839779f370a4753c0db26d6cf052203

[natanaeldgsantos]

I still have the same error too using databricks DBR 11.3 with spark 3.3.0 and DBR 12.2 with spark 3.3.2. im using pydeequ 1.2 deequ 2.0.4