关于内存:Python MemoryError – 有没有更有效的方法来处理我庞大的CSV文件?

Python MemoryError - Is there a more efficient way of working with my huge CSV file?

[使用Python3.3]我有一个巨大的CSV文件,其中包含XX万行并包含几列。 我想读取该文件,添加几个计算列并吐出几个"分段"的csv文件。 我在下面的代码中尝试了一个较小的测试文件,它完全符合我的要求。 但现在我正在加载原始的CSV文件(大约3.2 GB),我收到内存错误。 是否有更高效的内存编写以下代码?

请注意,我是Python的新手,因此可能有很多我不太了解的东西。

输入数据示例:

1
2
3
4
5
email               cc  nr_of_transactions  last_transaction_date   timebucket  total_basket
email1@email.com    us  2                   datetime value          1           20.29
email2@email.com    gb  3                   datetime value          2           50.84
email3@email.com    ca  5                   datetime value          3           119.12
...                 ... ...                 ...                     ...         ...

这是我的代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
import csv
import scipy.stats as stats
import itertools
from operator import itemgetter


def add_rankperc(filename):
    '''
    Function that calculates percentile rank of total basket value of a user (i.e. email) within a country. Next, it assigns the user to a rankbucket based on its percentile rank, using the following rules:
     Percentage rank between 75 and 100 -> top25
     Percentage rank between 25 and 74  -> mid50
     Percentage rank between 0 and 24   -> bottom25
    '''


    # Defining headers for ease of use/DictReader
    headers = ['email', 'cc', 'nr_transactions', 'last_transaction_date', 'timebucket', 'total_basket']
    groups = []

    with open(filename, encoding='utf-8', mode='r') as f_in:
        # Input file is tab-separated, hence dialect='excel-tab'
        r = csv.DictReader(f_in, dialect='excel-tab', fieldnames=headers)
        # DictReader reads all dict values as strings, converting total_basket to a float
        dict_list = []
        for row in r:
            row['total_basket'] = float(row['total_basket'])
            # Append row to a list (of dictionaries) for further processing
            dict_list.append(row)

    # Groupby function on cc and total_basket
    for key, group in itertools.groupby(sorted(dict_list, key=itemgetter('cc', 'total_basket')), key=itemgetter('cc')):
        rows = list(group)
        for row in rows:
            # Calculates the percentile rank for each value for each country
            row['rankperc'] = stats.percentileofscore([row['total_basket'] for row in rows], row['total_basket'])
            # Percentage rank between 75 and 100 -> top25
            if 75 <= row['rankperc'] <= 100:
                row['rankbucket'] = 'top25'
            # Percentage rank between 25 and 74 -> mid50
            elif 25 <= row['rankperc'] < 75:
                row['rankbucket'] = 'mid50'
            # Percentage rank between 0 and 24 -> bottom25
            else:
                row['rankbucket'] = 'bottom25'
            # Appending all rows to a list to be able to return it and use it in another function
            groups.append(row)
    return groups


def filter_n_write(data):
    '''
    Function takes input data, groups by specified keys and outputs only the e-mail addresses to csv files as per the respective grouping.
    '''


    # Creating group iterator based on keys
    for key, group in itertools.groupby(sorted(data, key=itemgetter('timebucket', 'rankbucket')), key=itemgetter('timebucket', 'rankbucket')):
        # List comprehension to create a list of lists of email addresses. One row corresponds to the respective combination of grouping keys.
        emails = list([row['email'] for row in group])
        # Dynamically naming output file based on grouping keys
        f_out = 'output-{}-{}.csv'.format(key[0], key[1])
        with open(f_out, encoding='utf-8', mode='w') as fout:
            w = csv.writer(fout, dialect='excel', lineterminator='
'
)
            # Writerows using list comprehension to write each email in emails iterator (i.e. one address per row). Wrapping email in brackets to write full address in one cell.
            w.writerows([email] for email in emails)

filter_n_write(add_rankperc('infile.tsv'))

提前致谢!


pandas库(http://pandas.pydata.org/)具有非常好的快速CSV读取功能(http://pandas.pydata.org/pandas-docs/stable/io.html#io-read-csv-表)。作为额外的奖励,您将数据作为numpy数组,使计算百分位数变得非常容易。
这个问题讨论了用pandas读取大块的CSV。


我同意Inbar Rose认为使用数据库功能会更好
攻击这个问题。假设我们需要回答你提出的问题,
虽然 - 我认为我们可以牺牲速度。

你可能在构造所有行的列表时内存不足'
字典。我们可以通过仅考虑行的子集来解决这个问题
一次。

这是第一步的代码 - 大致是你的add_rankperc函数:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
import csv
from scipy.stats import percentileofscore
from operator import itemgetter

# Run through the whole file once, saving each row to a file corresponding to
# its 'cc' column
cc_dict = {}
with open(input_path, encoding="utf-8", mode='r') as infile:
  csv_reader = csv.reader(infile, dialect="excel-tab")
  for row in csv_reader:
    cc = row[1]
    if cc not in cc_dict:
      intermediate_path ="intermediate_cc_{}.txt".format(cc)
      outfile = open(intermediate_path, mode='w', newline='')
      csv_writer = csv.writer(outfile)
      cc_dict[cc] = (intermediate_path, outfile, csv_writer)
    _ = cc_dict[cc][2].writerow(row)

# Close the output files
for cc in cc_dict.keys():
  cc_dict[cc][1].close()

# Run through the whole file once for each 'cc' value
for cc in cc_dict.keys():
  intermediate_path = cc_dict[cc][0]
  with open(intermediate_path, mode='r', newline='') as infile:
    csv_reader = csv.reader(infile)
    # Pick out all of the rows with the 'cc' value under consideration
    group = [row for row in csv_reader if row[1] == cc]
    # Get the 'total_basket' values for the group
    A_scores = [float(row[5]) for row in group]
    for row in group:
      # Compute this row's 'total_basket' score based on the rest of the
      # group's
      p = percentileofscore(A_scores, float(row[5]))
      row.append(p)
      # Categorize the score
      bucket = ("bottom25" if p < 25 else ("mid50" if p < 75 else"top100"))
      row.append(bucket)
  # Save the augmented rows to an intermediate file
  with open(output_path, mode='a', newline='') as outfile:
    csv_writer = csv.writer(outfile)
    csv_writer.writerows(group)

4600万行很多,所以这可能会很慢。我避免使用
csv模块的DictReader功能,只是为行编制索引
直接避免这种开销。我还计算了第一个参数
percentileofscores每组一次,而不是组中的每一行。

如果这样可行,那么我认为你可以对filter_n_write采用相同的想法
function - 运行生成的中间文件一次,挑选出来
(timebucket, rank)对。然后再次访问中间文件,一次
每一对。