对于我的一个数据分析管道，我最终生成了许多单独的 CSV 文件。我想转置它们，连接它们，然后再次转置它们。但是数据量很大，全部加载到内存中并不实用。
【问题讨论】：
连接两个 csv 文件中的数据行（如果您是这个意思）而不将它们都加载到内存中是一个相对简单且快速的操作：只需从每个文件中读取一行，然后将它们连接在一起，然后将其写入输出文件，重复直到所有输入数据用完为止。
在 csv 文件中转置数据而不是将整个内容读入内存本质上是一个慢得多的过程，因为它需要多次重新读取整个输入文件，每次只从一个文件中提取数据它包含的列。如果这是一个可以接受（或必要）的折衷方案，这基本上是使用内置 csv 模块完成的方法：
import csv
input_filename = 'input.csv'
output_filename = 'output.csv'
with open(output_filename, 'wb') as outputf:
writer = csv.writer(outputf)
with open(input_filename, 'rb') as inputf:
# determine number of columns in input file by counting those in its first row
# number of cols in input file determines number of rows in output file
numcols = len(csv.reader(inputf).next())
# read entire input file multiple times, extracting one column from each row
for col_index in xrange(numcols):
# write all of column data as a single row of the output file
inputf.seek(0) # rewind file for each pass
writer.writerow(tuple(row[col_index] for row in csv.reader(inputf)))
【讨论】：
struct 模块而不是csv 模块解析文件的问题似乎会更快。 然而，OP 对固定宽度字段只字未提。
以下代码模拟读取两个 csv 文件。第一个有两个 行
[1,2,1]
[3,4,1]
第二个
[7,8,2]
[9,10.2].
结果是两行
[1,2,1,7,8,2]
[3,4,1,9,10,2]
这就是你想要的吗？
def source1():
for i in [ [1,2, 1] ,[3,4, 1]] : yield i
def source2():
for i in [ [7,8,2] ,[9,10,2]] : yield i
def join(*sources):
while True:
row = []
for s in sources:
row.extend(s.next())
yield row
for row in join(source1(), source2()):
print row
在您的情况下，您必须用 csv 文件迭代器替换对 source1() 和 source2() 的调用。
【讨论】：
当字段具有固定宽度时，这是一个可行的解决方案：
import sys
import os
def main():
path_in = sys.argv[-1]
path_out = os.path.basename(path_in)+'.transposed'
with open(path_in) as fd_in:
line = fd_in.readline()
l = line.split()
field_width = int(len(line)/len(l))
file_size = os.path.getsize(path_in)
cols2 = rows1 = line_count = int(file_size/len(line))
rows2 = cols1 = len(l)
with open(path_in) as fd_in, open(path_out, 'w') as fd_out:
for row in range(rows2):
for col in range(cols2-1):
fd_in.seek(col*len(line)+row*field_width)
fd_out.write('{} '.format(fd_in.read(field_width-1)))
fd_in.seek((col+1)*len(line)+row*field_width)
fd_out.write('{}\n'.format(fd_in.read(field_width-1)))
return
if __name__ == '__main__':
main()
如果字段没有固定宽度，这是一个可行的解决方案：
import sys
import os
def main():
path_in = sys.argv[-1]
path_out = os.path.basename(path_in)+'.transposed'
separator = ' '
d_seek = {}
with open(path_in) as fd_in:
i = 0
while True:
tell = fd_in.tell()
if fd_in.readline() == '':
break
d_seek[i] = tell
i += 1
cols2 = rows1 = i
with open(path_in) as fd_in:
line = fd_in.readline()
rows2 = cols1 = len(line.split(separator))
del line
with open(path_in) as fd_in, open(path_out, 'w') as fd_out:
for row2 in range(rows2):
for row1 in range(rows1):
fd_in.seek(d_seek[row1])
j = 0
s = ''
while True:
char = fd_in.read(1)
j += 1
if char == separator or char == '\n':
break
s += char
d_seek[row1] += len(s)+1
if row1+1 < rows1:
fd_out.write('{} '.format(s))
else:
fd_out.write('{}\n'.format(s))
return
if __name__ == '__main__':
main()
【讨论】：
另一个简短的 Pythonic 解决方案。我用它来转置 15,000,000 x 12,000 的 CSV。它是快速而纯正的python。您需要做的其他一切都是微不足道的，这绝对是最难的部分。
Github 链接：https://gist.github.com/arose13/facfb91b609d453f3ad840417faa503a
def transpose_csv_out_of_core(csv_path, output_csv_path='transposed.csv', delimiter=','):
"""
On my laptop it can transpose at ~375,000 lines a sec
:param csv_path:
:param output_csv_path:
:param delimiter:
:return:
"""
import csv
transposed_iterator = zip(*csv.reader(open(csv_path)))
with open(output_csv_path, 'w') as out:
for row in transposed_iterator:
out.write(delimiter.join(row) + '\n')
【讨论】：
使用生成器，例如
from itertools import izip
file1 = open("test", "r")
file2 = open("test2", "r")
def lazy(file):
for line in file:
#do something with the line
yield line
for lines in izip(lazy(file1), lazy(file2)):
print lines
http://wiki.python.org/moin/Generators
编辑：你可以使用 CSV 模块来解析它，我也意识到文件对象的 readlines() 方法不是惰性的，所以你必须使用 for line in file 模式。
【讨论】：