我的pyspark数据框中有500列…有些是字符串类型,有些是int值,有些是布尔型(100个布尔型列).
现在,所有布尔值列都有两个不同的级别-是和否,我想将其转换为1/0

对于字符串,我有三个值：passed,failed和null.
如何将这些空值替换为0？
fillna(0)仅适用于整数

 c1| c2 |    c3 |c4|c5..... |c500
yes| yes|passed |45....
No | Yes|failed |452....
Yes|No  |None   |32............

当我做

df.replace(yes,1)

我收到以下错误：

ValueError: Mixed type replacements are not supported

解决方法:

对于字符串,我有三个值：passed,failed和null.如何将这些空值替换为0？ fillna(0)仅适用于整数

一,导入时和点亮

from pyspark.sql.functions import when, lit

假设您的DataFrame有这些列

# Reconstructing my DataFrame based on your assumptions
# cols are Columns in the DataFrame
cols = ['name', 'age', 'col_with_string']

# Similarly the values
vals = [
     ('James', 18, 'passed'),
     ('Smith', 15, 'passed'),
     ('Albie', 32, 'failed'),
     ('Stacy', 33, None),
     ('Morgan', 11, None),
     ('Dwight', 12, None),
     ('Steve', 16, 'passed'), 
     ('Shroud', 22, 'passed'),
     ('Faze', 11,'failed'),
     ('Simple', 13, None)
]

# This will create a DataFrame using 'cols' and 'vals'
# spark is an object of SparkSession
df = spark.createDataFrame(vals, cols)

# We have the following DataFrame
df.show()

+------+---+---------------+
|  name|age|col_with_string|
+------+---+---------------+
| James| 18|         passed|
| Smith| 15|         passed|
| Albie| 32|         failed|
| Stacy| 33|           null|
|Morgan| 11|           null|
|Dwight| 12|           null|
| Steve| 16|         passed|
|Shroud| 22|         passed|
|  Faze| 11|         failed|
|Simple| 13|           null|
+------+---+---------------+

您可以使用：

> withColumn()-指定要使用的列.
> isNull()-当属性评估为null时,评估结果为true的过滤器
> lit()-为文字创建一列
> when(),否则()-用于检查有关列的条件

我可以将具有null的值替换为0

df = df.withColumn('col_with_string', when(df.col_with_string.isNull(), 
lit('0')).otherwise(df.col_with_string))

# We have replaced nulls with a '0'
df.show()

+------+---+---------------+
|  name|age|col_with_string|
+------+---+---------------+
| James| 18|         passed|
| Smith| 15|         passed|
| Albie| 32|         failed|
| Stacy| 33|              0|
|Morgan| 11|              0|
|Dwight| 12|              0|
| Steve| 16|         passed|
|Shroud| 22|         passed|
|  Faze| 11|         failed|
|Simple| 13|              0|
+------+---+---------------+

问题的第1部分：是/否布尔值-您提到过,有100列布尔值.为此,我通常使用更新后的值来重建表,或者创建UDF返回1或0(表示是或否).

我将另外两列can_vote和can_lotto添加到DataFrame(df)

df = df.withColumn("can_vote", col('Age') >= 18)
df = df.withColumn("can_lotto", col('Age') > 16) 

# Updated DataFrame will be
df.show()

+------+---+---------------+--------+---------+
|  name|age|col_with_string|can_vote|can_lotto|
+------+---+---------------+--------+---------+
| James| 18|         passed|    true|     true|
| Smith| 15|         passed|   false|    false|
| Albie| 32|         failed|    true|     true|
| Stacy| 33|              0|    true|     true|
|Morgan| 11|              0|   false|    false|
|Dwight| 12|              0|   false|    false|
| Steve| 16|         passed|   false|    false|
|Shroud| 22|         passed|    true|     true|
|  Faze| 11|         failed|   false|    false|
|Simple| 13|              0|   false|    false|
+------+---+---------------+--------+---------+

假设您具有与can_vote和can_lotto相似的列(布尔值为Yes / No)

您可以使用以下代码行来获取具有布尔类型的DataFrame中的列

col_with_bool = [item[0] for item in df.dtypes if item[1].startswith('boolean')]

这将返回一个列表

['can_vote', 'can_lotto']

您可以创建一个UDF并为这种类型的列表中的每一列进行迭代,并使用1(是)或0(否)点亮每个列.

供参考,请参考以下链接

> isNull()：https://spark.apache.org/docs/2.2.0/api/java/org/apache/spark/sql/sources/IsNull.html
>点亮,时间：https://spark.apache.org/docs/1.6.2/api/java/org/apache/spark/sql/functions.html