如何使用 python 或 scala 将非标准 csv 文件读入数据帧

Question

我在下面有一个数据集采样，可以用python或scala处理：

FWD,13032009:09:01,10.56| FWD,13032009:10:53,11.23| FWD,13032009:15:40,23.20
SPOT,13032009:09:04,11.56| FWD,13032009:11:45,11.23| SPOT,13032009:12:30,23.20
FWD,13032009:08:01,10.56| SPOT,13032009:12:30,11.23| FWD,13032009:13:20,23.20| FWD,13032009:14:340,56.00
FWD,13032009:08:01,10.56| SPOT,13032009:12:30,11.23| FWD,13032009:13:20,23.20

每行都要拆分成多个较小的字符串，可以进一步拆分。

我正在寻找一种生成RDD或数据帧的有效方法，其中包含以下内容：

FWD,13032009:09:01,10.56 
FWD,13032009:10:53,11.23
FWD,13032009:15:40,23.20
SPOT,13032009:09:04,11.56
FWD,13032009:11:45,11.23
SPOT,13032009:12:30,23.20
FWD,13032009:08:01,10.56
SPOT,13032009:12:30,11.23
FWD,13032009:13:20,23.20
FWD,13032009:14:340,56.00
FWD,13032009:08:01,10.56
SPOT,13032009:12:30,11.23
FWD,13032009:13:20,23.20

请注意，效率越高越好，因为生产中的总行数可能高达百万

谢谢。

Answer 1

假设您正在从 csv 文件中读取，您可以将每一行读取到一个列表中。平展值，然后将它们作为单独的行进行处理。

将文件读入列表- 100 万行不应该太多，无法处理：

import csv
import itertools
import pandas as pd
with open('test.csv','r') as f:
reader = csv.reader(f, delimiter = '|')
rows = list(reader)

从单个列表中展平和拆分- Python 标准库中的优秀itertools库返回一个生成器，该生成器有助于内存且高效。

flat_rows = itertools.chain.from_iterable(rows)
list_rows = [i.strip().split(',') for i in flat_rows]

嵌套列表list_rows现在为您提供了一个干净且格式化的列表，如果您想创建dataframe，可以将其发送给pandas。

list_rows
>>
[['FWD', '13032009:09:01', '10.56'],
['FWD', '13032009:10:53', '11.23'],
['FWD', '13032009:15:40', '23.20'],
['SPOT', '13032009:09:04', '11.56'],
['FWD', '13032009:11:45', '11.23'],
['SPOT', '13032009:12:30', '23.20'],
['FWD', '13032009:08:01', '10.56'],
['SPOT', '13032009:12:30', '11.23'],
['FWD', '13032009:13:20', '23.20'],
['FWD', '13032009:14:340', '56.00'],
['FWD', '13032009:08:01', '10.56'],
['SPOT', '13032009:12:30', '11.23'],
['FWD', '13032009:13:20', '23.20']]
df = pd.DataFrame(list_rows)

Answer 2

Python 解决方案：如果以字符串形式获取文本，则可以使用换行符 (n(replace()| 序列，然后将其读取为数据帧：

import pandas as pd
from io import StringIO
data_set = """FWD,13032009:09:01,10.56| FWD,13032009:10:53,11.23| FWD,13032009:15:40,23.20
SPOT,13032009:09:04,11.56| FWD,13032009:11:45,11.23| SPOT,13032009:12:30,23.20
FWD,13032009:08:01,10.56| SPOT,13032009:12:30,11.23| FWD,13032009:13:20,23.20| FWD,13032009:14:340,56.00
FWD,13032009:08:01,10.56| SPOT,13032009:12:30,11.23| FWD,13032009:13:20,23.20
"""
data_set *= 100000  # Make it over a million elements to ensure performance is adequate
data_set = data_set.replace("| ", "n")
data_set_stream = StringIO(data_set)  # Pandas needs to read a file-like object, so need to turn our string into a buffer
df = pd.read_csv(data_set_stream)
print(df)  # df is our desired DataFrame

Answer 3

如果你有兴趣，这里是Scala的方式，

val rdd1 = sc.parallelize(List("FWD,13032009:09:01,10.56| FWD,13032009:10:53,11.23| FWD,13032009:15:40,23.20", "SPOT,13032009:09:04,11.56| FWD,13032009:11:45,11.23| SPOT,13032009:12:30,23.20","FWD,13032009:08:01,10.56| SPOT,13032009:12:30,11.23| FWD,13032009:13:20,23.20| FWD,13032009:14:340,56.00","FWD,13032009:08:01,10.56| SPOT,13032009:12:30,11.23| FWD,13032009:13:20,23.20"))

val rdd2 = rdd1.flatMap(l => l.replaceAll(" ","").split("\|")) val rds = rdd2.toDS

val df = spark.read.csv(rds)

df.show(false) +----+---------------+-----+ |_c0 |_c1 |_c2 | +----+---------------+-----+ |FWD |13032009:09:01 |10.56| |FWD |13032009:10:53 |11.23| |FWD |13032009:15:40 |23.20| |SPOT|13032009:09:04 |11.56| |FWD |13032009:11:45 |11.23| |SPOT|13032009:12:30 |23.20| |FWD |13032009:08:01 |10.56| |SPOT|13032009:12:30 |11.23| |FWD |13032009:13:20 |23.20| |FWD |13032009:14:340|56.00| |FWD |13032009:08:01 |10.56| |SPOT|13032009:12:30 |11.23| |FWD |13032009:13:20 |23.20| +----+---------------+-----+