在 PySpark 中的多列上应用 MinMaxScaler

Question

我想将MinMaxScalarPySpark 应用于 PySpark 数据框的多列df。到目前为止，我只知道如何将它应用于单个列，例如x.

from pyspark.ml.feature import MinMaxScaler

pdf = pd.DataFrame({'x':range(3), 'y':[1,2,5], 'z':[100,200,1000]})
df = spark.createDataFrame(pdf)

scaler = MinMaxScaler(inputCol="x", outputCol="x")
scalerModel = scaler.fit(df)
scaledData = scalerModel.transform(df)

如果我有 100 列怎么办？有没有办法对 PySpark 中的许多列进行最小-最大缩放？

更新：

另外，如何应用MinMaxScalar整数或双精度值？它引发以下错误：

java.lang.IllegalArgumentException: requirement failed: Column length must be of type struct<type:tinyint,size:int,indices:array<int>,values:array<double>> but was actually int.

Answer 1

问题 1：

如何更改您的示例以正常运行。您需要准备数据作为转换器工作的向量。

from pyspark.ml.feature import MinMaxScaler
from pyspark.ml import Pipeline
from pyspark.ml.linalg import VectorAssembler

pdf = pd.DataFrame({'x':range(3), 'y':[1,2,5], 'z':[100,200,1000]})
df = spark.createDataFrame(pdf)

assembler = VectorAssembler(inputCols=["x"], outputCol="x_vec")
scaler = MinMaxScaler(inputCol="x_vec", outputCol="x_scaled")
pipeline = Pipeline(stages=[assembler, scaler])
scalerModel = pipeline.fit(df)
scaledData = scalerModel.transform(df)

问题2：

要在多个列上运行 MinMaxScaler，您可以使用接收带有列表理解准备的转换列表的管道：

from pyspark.ml import Pipeline
from pyspark.ml.feature import MinMaxScaler
columns_to_scale = ["x", "y", "z"]
assemblers = [VectorAssembler(inputCols=[col], outputCol=col + "_vec") for col in columns_to_scale]
scalers = [MinMaxScaler(inputCol=col + "_vec", outputCol=col + "_scaled") for col in columns_to_scale]
pipeline = Pipeline(stages=assemblers + scalers)
scalerModel = pipeline.fit(df)
scaledData = scalerModel.transform(df)

在官方文档中查看此示例管道。

最终，您将得到以下格式的结果：

>>> scaledData.printSchema() 
root
 |-- x: long (nullable = true)
 |-- y: long (nullable = true)
 |-- z: long (nullable = true)
 |-- x_vec: vector (nullable = true)
 |-- y_vec: vector (nullable = true)
 |-- z_vec: vector (nullable = true)
 |-- x_scaled: vector (nullable = true)
 |-- y_scaled: vector (nullable = true)
 |-- z_scaled: vector (nullable = true)

>>> scaledData.show()
+---+---+----+-----+-----+--------+--------+--------+--------------------+
|  x|  y|   z|x_vec|y_vec|   z_vec|x_scaled|y_scaled|            z_scaled|
+---+---+----+-----+-----+--------+--------+--------+--------------------+
|  0|  1| 100|[0.0]|[1.0]| [100.0]|   [0.0]|   [0.0]|               [0.0]|
|  1|  2| 200|[1.0]|[2.0]| [200.0]|   [0.5]|  [0.25]|[0.1111111111111111]|
|  2|  5|1000|[2.0]|[5.0]|[1000.0]|   [1.0]|   [1.0]|               [1.0]|
+---+---+----+-----+-----+--------+--------+--------+--------------------+

额外的后处理：

您可以通过一些后处理以原始名称恢复列。例如：

from pyspark.sql import functions as f
names = {x + "_scaled": x for x in columns_to_scale}
scaledData = scaledData.select([f.col(c).alias(names[c]) for c in names.keys()])

输出将是：

scaledData.show()
+------+-----+--------------------+
|     y|    x|                   z|
+------+-----+--------------------+
| [0.0]|[0.0]|               [0.0]|
|[0.25]|[0.5]|[0.1111111111111111]|
| [1.0]|[1.0]|               [1.0]|
+------+-----+--------------------+

Answer 2

您可以将单个 MinMaxScaler 实例用于一组“矢量组装”功能，而不是为要转换的每一列创建一个 MinMaxScaler（在本例中为缩放）。

from pyspark.ml.feature import MinMaxScaler
from pyspark.ml.feature import VectorAssembler

#1. Your original dataset
#pdf = pd.DataFrame({'x':range(3), 'y':[1,2,5], 'z':[100,200,1000]})
#df = spark.createDataFrame(pdf)

df = spark.createDataFrame([(0, 10.0, 0.1), (1, 1.0, 0.20), (2, 1.0, 0.9)],["x", "y", "z"])

df.show()
+---+----+---+
|  x|   y|  z|
+---+----+---+
|  0|10.0|0.1|
|  1| 1.0|0.2|
|  2| 1.0|0.9|
+---+----+---+

#2. Vector assembled set of features 
# (assemble only the columns you want to MinMax Scale)
assembler = VectorAssembler(inputCols=["x", "y", "z"], 
outputCol="features")
output = assembler.transform(df)

output.show()

+---+----+---+--------------+
|  x|   y|  z|      features|
+---+----+---+--------------+
|  0|10.0|0.1|[0.0,10.0,0.1]|
|  1| 1.0|0.2| [1.0,1.0,0.2]|
|  2| 1.0|0.9| [2.0,1.0,0.9]|
+---+----+---+--------------+

#3. Applying MinMaxScaler to your assembled features 
scaler = MinMaxScaler(inputCol="features", outputCol="scaledFeatures")
# rescale each feature to range [min, max].
scaledData = scaler.fit(output).transform(output)
scaledData.show()

+---+----+---+--------------+---------------+
|  x|   y|  z|      features| scaledFeatures|
+---+----+---+--------------+---------------+
|  0|10.0|0.1|[0.0,10.0,0.1]|  [0.0,1.0,0.0]|
|  1| 1.0|0.2| [1.0,1.0,0.2]|[0.5,0.0,0.125]|
|  2| 1.0|0.9| [2.0,1.0,0.9]|  [1.0,0.0,1.0]|
+---+----+---+--------------+---------------+

希望这可以帮助。