使用 StandardScaler 时的 SparseVector 与 DenseVector答案

【问题标题】：SparseVector vs DenseVector when using StandardScaler使用 StandardScaler 时的 SparseVector 与 DenseVector
【发布时间】：2016-12-21 09:56:32
【问题描述】：

我正在使用以下代码来规范化 PySpark DataFrame

from pyspark.ml.feature import StandardScaler, VectorAssembler
from pyspark.ml import Pipeline

cols = ["a", "b", "c"]
df = spark.createDataFrame([(1, 0, 3), (2, 3, 2), (1, 3, 1), (3, 0, 3)], cols)

Pipeline(stages=[
    VectorAssembler(inputCols=cols, outputCol='features'), 
    StandardScaler(withMean=True, inputCol='features', outputCol='scaledFeatures')
]).fit(df).transform(df).select(cols + ['scaledFeatures']).head()

这给出了预期的结果：

Row(a=1, b=0, c=3, scaledFeatures=DenseVector([-0.7833, -0.866, 0.7833]))

但是，当我在从 parquet 文件加载的（大得多的）数据集上运行管道时，我收到以下异常：

16/12/21 09:47:50 WARN TaskSetManager: Lost task 0.0 in stage 60.0 (TID 6370, 10.231.153.67): org.apache.spark.SparkException: Failed to execute user defined function($anonfu
n$2: (vector) => vector)
        at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply2_2$(Unknown Source)
        at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
        at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
        at scala.collection.Iterator$$anon$11.next(Iterator.scala:409)
        at scala.collection.Iterator$$anon$11.next(Iterator.scala:409)
        at org.apache.spark.api.python.SerDeUtil$AutoBatchedPickler.next(SerDeUtil.scala:121)
        at org.apache.spark.api.python.SerDeUtil$AutoBatchedPickler.next(SerDeUtil.scala:112)
        at scala.collection.Iterator$class.foreach(Iterator.scala:893)
        at org.apache.spark.api.python.SerDeUtil$AutoBatchedPickler.foreach(SerDeUtil.scala:112)
        at org.apache.spark.api.python.PythonRDD$.writeIteratorToStream(PythonRDD.scala:504)
        at org.apache.spark.api.python.PythonRunner$WriterThread$$anonfun$run$3.apply(PythonRDD.scala:328)
        at org.apache.spark.util.Utils$.logUncaughtExceptions(Utils.scala:1877)
        at org.apache.spark.api.python.PythonRunner$WriterThread.run(PythonRDD.scala:269)
Caused by: java.lang.IllegalArgumentException: Do not support vector type class org.apache.spark.mllib.linalg.SparseVector
        at org.apache.spark.mllib.feature.StandardScalerModel.transform(StandardScaler.scala:160)
        at org.apache.spark.ml.feature.StandardScalerModel$$anonfun$2.apply(StandardScaler.scala:167)
        at org.apache.spark.ml.feature.StandardScalerModel$$anonfun$2.apply(StandardScaler.scala:167)
        ... 13 more

我注意到这里的 VectorAssembler 已将我的列转换为 mllib.linalg.SparseVector 而不是第一种情况中使用的 DenseVector。

有什么办法可以解决这个问题吗？

【问题讨论】：

您使用的是哪个版本的 spark？
火花 2.0.1。很确定这个答案stackoverflow.com/questions/35844330/… 是关键。目前正在尝试将 SparseVector 转换为 DenseVector，但这也不是直截了当的。
“b = DenseVector(a.toArray())”不是一个直接的解决方案吗？
Spark 还是个新手。我正在弄清楚如何将该转换应用于数据框中的列。 udf 是最好的选择吗？例如。 asDense = udf(lambda s: DenseVector(s.toArray()), VectorUDT()) df = df.withColumn('features', asDense(df.features))
也可以将其添加为管道中的转换，但我不确定如何添加任意转换......

标签： apache-spark pyspark spark-dataframe apache-spark-mllib pyspark-sql

【解决方案1】：

我注意到您希望将其创建为自定义转换以将其直接包含在您的管道中。

这应该会为你做到这一点。

from pyspark import keyword_only  
from pyspark.ml.pipeline import Transformer  
from pyspark.ml.param.shared import HasInputCol, HasOutputCol
from pyspark.ml.linalg import SparseVector, DenseVector, VectorUDT
from pyspark.sql.functions import udf


class AsDenseTransformer(Transformer, HasInputCol, HasOutputCol):  
    @keyword_only
    def __init__(self, inputCol=None, outputCol=None):
        super(AsDenseTransformer, self).__init__()
        kwargs = self.__init__._input_kwargs
        self.setParams(**kwargs)

    @keyword_only
    def setParams(self, inputCol=None, outputCol=None):
        kwargs = self.setParams._input_kwargs
        return self._set(**kwargs)

    def _transform(self, dataset):
        out_col = self.getOutputCol()
        in_col = dataset[self.getInputCol()]

        asDense = udf(lambda s: DenseVector(s.toArray()), VectorUDT()) 

        return dataset.withColumn(out_col,  asDense(in_col))

一旦你定义了它，你可以将它初始化为一个转换，在向量汇编器之后包含在你的管道中。

Pipeline(stages=[
    VectorAssembler(inputCols=cols, outputCol='features'),
    AsDenseTransformer(inputCol='features', outputCol='features'),
    StandardScaler(withMean=True, inputCol='features', outputCol='scaledFeatures')
]).fit(df).transform(df).select(cols + ['scaledFeatures']).head()

【讨论】：