我有一个从CSV文件读取的pyspark数据帧,该文件有一个包含十六进制值的值列。
| date | part | feature | value" |
|----------|-------|---------|--------------|
| 20190503 | par1 | feat2 | 0x0 |
| 20190503 | par1 | feat3 | 0x01 |
| 20190501 | par2 | feat4 | 0x0f32 |
| 20190501 | par5 | feat9 | 0x00 |
| 20190506 | par8 | feat2 | 0x00f45 |
| 20190507 | par1 | feat6 | 0x0e62300000 |
| 20190501 | par11 | feat3 | 0x000000000 |
| 20190501 | par21 | feat5 | 0x03efff |
| 20190501 | par3 | feat9 | 0x000 |
| 20190501 | par6 | feat5 | 0x000000 |
| 20190506 | par5 | feat8 | 0x034edc45 |
| 20190506 | par8 | feat1 | 0x00000 |
| 20190508 | par3 | feat6 | 0x00000000 |
| 20190503 | par4 | feat3 | 0x0c0deffe21 |
| 20190503 | par6 | feat4 | 0x0000000000 |
| 20190501 | par3 | feat6 | 0x0123fe |
| 20190501 | par7 | feat4 | 0x00000d0 |
要求是删除在value列中包含类似于0x0、0x00、0x000等的值的行,这些值的计算结果为十进制0(零)。“0x”之后的0的数量在整个数据帧中不同。通过模式匹配删除是我尝试的方法,但我没有成功。
myFile = sc.textFile("file.txt")
header = myFile.first()
fields = [StructField(field_name, StringType(), True) for field_name in header.split(',')]
myFile_header = myFile.filter(lambda l: "date" in l)
myFile_NoHeader = myFile.subtract(myFile_header)
myFile_df = myFile_NoHeader.map(lambda line: line.split(",")).toDF(schema)
## this is the pattern match I tried
result = myFile_df.withColumn('Test', regexp_extract(col('value'), '(0x)(0\1*\1*)',2 ))
result.show()
我使用的另一种方法是使用udf:
def convert_value(x):
return int(x,16)
在pyspark中使用这个udf给我
ValueError:基数为16的int()的文本无效:值为
发布于 2019-05-13 06:14:14
我不太理解您的正则表达式,但是当您想要匹配所有包含0x0 (+任意数量的零)的字符串时,您可以使用^0x0+$
。使用正则表达式的过滤可以通过rlike实现,而代字号将否定匹配。
l = [('20190503', 'par1', 'feat2', '0x0'),
('20190503', 'par1', 'feat3', '0x01'),
('20190501', 'par2', 'feat4', '0x0f32'),
('20190501', 'par5', 'feat9', '0x00'),
('20190506', 'par8', 'feat2', '0x00f45'),
('20190507', 'par1', 'feat6', '0x0e62300000'),
('20190501', 'par11', 'feat3', '0x000000000'),
('20190501', 'par21', 'feat5', '0x03efff'),
('20190501', 'par3', 'feat9', '0x000'),
('20190501', 'par6', 'feat5', '0x000000'),
('20190506', 'par5', 'feat8', '0x034edc45'),
('20190506', 'par8', 'feat1', '0x00000'),
('20190508', 'par3', 'feat6', '0x00000000'),
('20190503', 'par4', 'feat3', '0x0c0deffe21'),
('20190503', 'par6', 'feat4', '0x0000000000'),
('20190501', 'par3', 'feat6', '0x0123fe'),
('20190501', 'par7', 'feat4', '0x00000d0')]
columns = ['date', 'part', 'feature', 'value']
df=spark.createDataFrame(l, columns)
expr = "^0x0+$"
df.filter(~ df["value"].rlike(expr)).show()
输出:
+--------+-----+-------+------------+
| date| part|feature| value|
+--------+-----+-------+------------+
|20190503| par1| feat3| 0x01|
|20190501| par2| feat4| 0x0f32|
|20190506| par8| feat2| 0x00f45|
|20190507| par1| feat6|0x0e62300000|
|20190501|par21| feat5| 0x03efff|
|20190506| par5| feat8| 0x034edc45|
|20190503| par4| feat3|0x0c0deffe21|
|20190501| par3| feat6| 0x0123fe|
|20190501| par7| feat4| 0x00000d0|
+--------+-----+-------+------------+
https://stackoverflow.com/questions/56103432
复制相似问题