如何在没有 pandas 的情况下读取、格式化、排序和保存 csv 文件

Posted 2023-02-23

【中文标题】如何在没有 pandas 的情况下读取、格式化、排序和保存 csv 文件

【英文标题】：How to read, format, sort, and save a csv file, without pandas

【发布时间】：2021-01-12 22:29:49

【问题描述】：

给定文件test.csv中的以下示例数据

27-Mar-12,8.25,8.35,8.17,8.19,9801989
26-Mar-12,8.16,8.25,8.12,8.24,8694416
23-Mar-12,8.05,8.12,7.95,8.09,8149170

如何在不使用pandas 的情况下解析此文件？

打开文件 将日期列格式化为datetime日期格式字符串 按第 0 列、日期列对所有行进行排序 保存回同一个文件，并带有日期列的标题

使用pandas，这可以通过一行（长）代码完成，不包括导入。 需要注意的是，如果不使用date_parser，使用parse_date可能会很慢。

import pandas as pd

(pd.read_csv('test.csv', header=None, parse_dates=[0], date_parser=lambda t: pd.to_datetime(t, format='%d-%b-%y'))
 .rename(columns=0: 'date')
 .sort_values('date')
 .to_csv('test.csv', index=False))

预期形式

date,1,2,3,4,5
2012-03-23,8.05,8.12,7.95,8.09,8149170
2012-03-26,8.16,8.25,8.12,8.24,8694416
2012-03-27,8.25,8.35,8.17,8.19,9801989

编写此问题和答案是为了填补 Stack Overflow 上的知识内容空白。 使用pandas 来完成这项任务非常简单。 在没有pandas 的情况下，想出所有必要的部分来创建一个完整的解决方案是非常困难的。 这应该对任何对此任务感兴趣的人以及禁止使用pandas 的学生有所帮助。 我不介意看到 numpy 的解决方案，但问题的主要点是仅使用标准库中的包来完成此任务。

三个答案都是可以接受的问题的解决方案。

【参考方案1】：

到目前为止，pandas 是解析和清理文件的更简单工具。 需要 1 行 pandas，占用 11 行代码，还需要 for-loop。 这需要以下包和函数 csv & datetime Methods of File Objects: .seek & .truncate Sorting: How To 最初，list() 用于解压缩 csv.reader 对象，但已被删除，以更新日期值，同时迭代 reader。 可以向sorted 提供自定义键函数以自定义排序顺序，但我看不到从lambda 表达式返回值的方法。 最初使用 key=lambda row: datetime.strptime(row[0], '%Y-%m-%d')，但已被删除，因为更新的日期列不包含月份名称。 如果日期列包含月份名称，则如果没有自定义排序键，它将无法正确排序。

import csv
from datetime import datetime

# open the file for reading and writing
with open('test1.csv', mode='r+', newline='') as f:

   # create a reader and writer opbject
    reader, writer = csv.reader(f), csv.writer(f)
    
    data = list()
    # iterate through the reader and update column 0 to a datetime date string
    for row in reader:

        # update column 0 to a datetime date string
        row[0] = datetime.strptime(row[0], "%d-%b-%y").date().isoformat()
        
        # append the row to data
        data.append(row)

    # sort all of the rows, based on date, with a lambda expression
    data = sorted(data, key=lambda row: row[0])

    # change the stream position to the given byte offset
    f.seek(0)

    # truncate the file size
    f.truncate()

    # add a header to data
    data.insert(0, ['date', 1, 2, 3, 4, 5])

    # write data to the file
    writer.writerows(data)

更新test.csv

date,1,2,3,4,5
2012-03-23,8.05,8.12,7.95,8.09,8149170
2012-03-26,8.16,8.25,8.12,8.24,8694416
2012-03-27,8.25,8.35,8.17,8.19,9801989

%time测试

import pandas
import pandas_datareader as web

# test data with 1M rows
df = web.DataReader(ticker, data_source='yahoo', start='1980-01-01', end='2020-09-27').drop(columns=['Adj Close']).reset_index().sort_values('High', ascending=False)
df.Date = df.Date.dt.strftime('%d-%b-%y')
df = pd.concat([df]*100)
df.to_csv('test.csv', index=False, header=False)

测试

# pandas test with date_parser
%time pandas_test('test.csv')
[out]:
Wall time: 17.9 s

# pandas test without the date_parser parameter
%time pandas_test('test.csv')
[out]:
Wall time: 1min 17s

# from Paddy Alton
%time paddy('test.csv')
[out]:
Wall time: 15.9 s

# from Trenton
%time trenton('test.csv')
[out]:
Wall time: 17.7 s

# from sammywemmy with functions updated to return the correct date format
%time sammy('test.csv')
[out]:
Wall time: 22.2 s
%time sammy2('test.csv')
[out]:
Wall time: 22.2 s

测试函数

from operator import itemgetter
import csv
import pandas as pd
from datetime import datetime

def pandas_test(file):
    (pd.read_csv(file, header=None, parse_dates=[0], date_parser=lambda t: pd.to_datetime(t, format='%d-%b-%y'))
     .rename(columns=0: 'date')
     .sort_values('date')
     .to_csv(file, index=False))


def trenton(file):
    with open(file, mode='r+', newline='') as f:
        reader, writer = csv.reader(f), csv.writer(f)
        data = list()
        for row in reader:
            row[0] = datetime.strptime(row[0], "%d-%b-%y").date().isoformat()
            data.append(row)
        data = sorted(data, key=lambda row: row[0])
        f.seek(0)
        f.truncate()
        data.insert(0, ['date', 1, 2, 3, 4, 5])
        writer.writerows(data)


def paddy(file):
    def format_date(date: str) -> str:
        formatted_date = datetime.strptime(date, "%d-%b-%y").date().isoformat()
        return formatted_date


    with open(file, "r") as f:
        lines = f.readlines()
        records = [[value for value in line.split(",")] for line in lines]
    for record in records:
        record[0] = format_date(record[0])
    sorted_records = sorted(records, key = lambda r: r[0])
    prepared_lines = [",".join(record).strip("\n") for record in sorted_records]
    field_names = "date,1,2,3,4,5"
    prepared_lines.insert(0, field_names)
    prepared_data = "\n".join(prepared_lines)
    with open(file, "w") as f:
        f.write(prepared_data)


def sammy(file):
    # updated with .date().isoformat() to return the correct format
    with open(file) as csvfile:
        fieldnames = ["date", 1, 2, 3, 4, 5]
        reader = csv.DictReader(csvfile, fieldnames=fieldnames)
        mapping = list(reader)
        mapping = [
            
                key: datetime.strptime(value, ("%d-%b-%y")).date().isoformat()
                     if key == "date" else value
                for key, value in entry.items()
            
            for entry in mapping
        ]

        mapping = sorted(mapping, key=itemgetter("date"))

    with open(file, mode="w", newline="") as csvfile:
        fieldnames = mapping[0].keys()
        writer = csv.DictWriter(csvfile, fieldnames=fieldnames)

        writer.writeheader()
        for row in mapping:
            writer.writerow(row)


def sammy2(file):
    # updated with .date().isoformat() to return the correct format
    with open(file) as csvfile:
        reader = csv.reader(csvfile, delimiter=",")
        mapping = dict(enumerate(reader))
        num_of_cols = len(mapping[0])
        fieldnames = ["date" if n == 0 else n
                      for n in range(num_of_cols)]
        mapping = [
                   [ datetime.strptime(val, "%d-%b-%y").date().isoformat()
                     if ind == 0 else val
                     for ind, val in enumerate(value)
                   ]
                 for key, value in mapping.items()
                  ]

        mapping = sorted(mapping, key=itemgetter(0))

    with open(file, mode="w", newline="") as csvfile:
        csvwriter = csv.writer(csvfile, delimiter=",")
        csvwriter.writerow(fieldnames)
        for row in mapping:
            csvwriter.writerow(row)

【讨论】：

我喜欢seek和truncate的用法。我已经看到了一些关于 SO 的问题，人们“只想保存”对 CSV 的更改，但我没有一个好的解决方案；我总是推荐'read-old，write-new，mv/rename old→new'。我已经收藏了，谢谢！

【参考方案2】：

正如 OP 所述，Pandas 让这一切变得简单；另一种方法是使用DictReader 和DictWriter 选项；它仍然比使用 Pandas 更冗长（这里的抽象之美，Pandas 为我们完成了繁重的工作）。

import csv
from datetime import datetime
from operator import itemgetter

with open("test.csv") as csvfile:
    fieldnames = ["date", 1, 2, 3, 4, 5]
    reader = csv.DictReader(csvfile, fieldnames=fieldnames)
    mapping = list(reader)
    mapping = [
        
            key: datetime.strptime(value, ("%d-%b-%y"))
                 if key == "date" else value
            for key, value in entry.items()
        
        for entry in mapping
    ]

    mapping = sorted(mapping, key=itemgetter("date"))

with open("test.csv", mode="w", newline="") as csvfile:
    fieldnames = mapping[0].keys()
    writer = csv.DictWriter(csvfile, fieldnames=fieldnames)

    writer.writeheader()
    for row in mapping:
        writer.writerow(row)

由于事先不知道字段名称，我们可以使用csvreader 和csvwriter 选项：

with open("test.csv") as csvfile:
    reader = csv.reader(csvfile, delimiter=",")
    mapping = dict(enumerate(reader))
    num_of_cols = len(mapping[0])
    fieldnames = ["date" if n == 0 else n
                  for n in range(num_of_cols)]
    mapping = [
               [ datetime.strptime(val, "%d-%b-%y")
                 if ind == 0 else val
                 for ind, val in enumerate(value)
               ]
             for key, value in mapping.items()
              ]

    mapping = sorted(mapping, key=itemgetter(0))

with open("test.csv", mode="w", newline="") as csvfile:
    csvwriter = csv.writer(csvfile, delimiter=",")
    csvwriter.writerow(fieldnames)
    for row in mapping:
        csvwriter.writerow(row)

【参考方案3】：

尽可能少地使用导入：

from datetime import datetime

def format_date(date: str) -> str:

    formatted_date = datetime.strptime(date, "%d-%b-%y").date().isoformat()

    return formatted_date


# read in the CSV
with open("test.csv", "r") as file:

    lines = file.readlines()

    records = [[value for value in line.split(",")] for line in lines]

# reformat the first field in each record
for record in records:
    record[0] = format_date(record[0])

# having formatted the dates, sort records by first (date) field:
sorted_records = sorted(records, key = lambda r: r[0])

# join values with commas once more, removing newline characters
prepared_lines = [",".join(record).strip("\n") for record in sorted_records]

# create a header row
field_names = "date,1,2,3,4,5"

# prepend the header row
prepared_lines.insert(0, field_names)

prepared_data = "\n".join(prepared_lines)

# write out the CSV
with open("test.csv", "w") as file:
    file.write(prepared_data)

【讨论】：

您可以跳过.strip("\n") 和"\n".join()，然后使用file.write(field_names+"\n")，然后使用file.writelines(prepared_lines)。

肯定有几种方法可以管理代码的读/写位，而且会更短。此外，回想起来，也许format_date 应该将记录作为输入并格式化其第一个元素以允许更优雅的formatted_records = list(map(format_date, records)) 代替循环。