我找到了一个似乎有效的解决方案，以防有人感兴趣。答案在于创建自定义约束而不是自定义分析器。以下是工作代码：

val myConstraint = Constraint.complianceConstraint(
    "my constraint",
    "`scrubbed` NOT IN ('ZZ', 'XX') AND `scrubbed` IS NOT NULL",
    (fraction:Double)=>fraction>=0.8,
    Some("`orig` NOT IN ('ZZ', 'XX') AND `orig` IS NOT NULL"),
    Some("no peeking")
)

val myVerificationResult: VerificationResult = { VerificationSuite()
    .onData(df)
    .addCheck(
        Check(CheckLevel.Error, "Review Check") 
            .addConstraint(myConstraint)
    )
    .run()
}
val result = checkResultsAsDataFrame(spark, myVerificationResult)
result.show(truncate=true)

结果与预期完全一致：

+------------+-----------+------------+--------------------+-----------------+--------------------+
|       check|check_level|check_status|          constraint|constraint_status|  constraint_message|
+------------+-----------+------------+--------------------+-----------------+--------------------+
|Review Check|      Error|       Error|ComplianceConstra...|          Failure|Value: 0.75 does ...|
+------------+-----------+------------+--------------------+-----------------+--------------------+

这不能只通过使用统计进行检查吗https://github.com/awslabs/deequ/blob/master/src/main/scala/com/amazon/deequ/checks/check.scala#l667 像这样的

Check(CheckLevel.Warning, "Statisfies TEST Constraint")      
      .satisfies("`scrubbed` NOT IN ('ZZ', 'XX') AND `scrubbed` IS NOT NULL",
                 "my constraint",
                 "fraction:Double",(fraction:Double)=>fraction>=0.8,
                  Some("..."))
      ))

我认为这是oob，而不是通过合规约束来定义，虽然这也是一个食品的想法，如果你有一个复杂的逻辑。