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

    https://github.com/apache/spark/pull/16776#discussion_r99999962
  
    --- Diff: 
sql/core/src/main/scala/org/apache/spark/sql/DataFrameStatFunctions.scala ---
    @@ -63,44 +63,49 @@ final class DataFrameStatFunctions private[sql](df: 
DataFrame) {
        *   Note that values greater than 1 are accepted but give the same 
result as 1.
        * @return the approximate quantiles at the given probabilities
        *
    -   * @note NaN values will be removed from the numerical column before 
calculation
    +   * @note null and NaN values will be removed from the numerical column 
before calculation
        *
        * @since 2.0.0
        */
       def approxQuantile(
           col: String,
           probabilities: Array[Double],
           relativeError: Double): Array[Double] = {
    -    StatFunctions.multipleApproxQuantiles(df.select(col).na.drop(),
    -      Seq(col), probabilities, relativeError).head.toArray
    +    val res = approxQuantile(Array(col), probabilities, relativeError)
    +    if (res != null) {
    +      res.head
    +    } else {
    +      null
    +    }
       }
     
       /**
        * Calculates the approximate quantiles of numerical columns of a 
DataFrame.
    -   * @see [[DataFrameStatsFunctions.approxQuantile(col:Str* 
approxQuantile]] for
    -   *     detailed description.
    +   * @see `DataFrameStatsFunctions.approxQuantile` for detailed 
description.
        *
    -   * Note that rows containing any null or NaN values values will be 
removed before
    -   * calculation.
        * @param cols the names of the numerical columns
        * @param probabilities a list of quantile probabilities
        *   Each number must belong to [0, 1].
        *   For example 0 is the minimum, 0.5 is the median, 1 is the maximum.
    -   * @param relativeError The relative target precision to achieve (>= 0).
    +   * @param relativeError The relative target precision to achieve 
(greater or equal to 0).
        *   If set to zero, the exact quantiles are computed, which could be 
very expensive.
        *   Note that values greater than 1 are accepted but give the same 
result as 1.
        * @return the approximate quantiles at the given probabilities of each 
column
        *
    -   * @note Rows containing any NaN values will be removed before 
calculation
    +   * @note Rows containing any null or NaN values will be removed before 
calculation
        *
        * @since 2.2.0
        */
       def approxQuantile(
           cols: Array[String],
           probabilities: Array[Double],
           relativeError: Double): Array[Array[Double]] = {
    -    StatFunctions.multipleApproxQuantiles(df.select(cols.map(col): 
_*).na.drop(), cols,
    -      probabilities, relativeError).map(_.toArray).toArray
    +    try {
    +      StatFunctions.multipleApproxQuantiles(df.select(cols.map(col): 
_*).na.drop(), cols,
    --- End diff --
    
    We will drop the whole row if any column has `null` or `NaN`. For example,
    ```Scala
        Seq[(java.lang.Long, java.lang.Double)]((null, 1.23), (3L, null), (4L, 
3.45))
          .toDF("a", "b").na.drop().show()
    ```
    
    That means, users could get different results. It depends on which API they 
used.
    ```Scala
    df.stat.approxQuantile("col1", Array(q1), epsilon)
    df.stat.approxQuantile("col2", Array(q1), epsilon)
    ```
    ```Scala
    df.stat.approxQuantile(Array("col1", "col2"), Array(q1), epsilon)
    ```
    
    I am wondering if this is the expected behavior? 


---
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