我不认为pd.rad_csv有一个很好的方法来做你想做的事情，一个选择是使用how to efficiently get line count的答案，并在代码中使用它

# file_name = 'myfile.csv'
num_lines = sum(1 for _ in open(file_name))

rows_wanted = [x if x>= 0 else num_lines + x for x in rows_wanted]
pd.read_csv('titanic.csv', skiprows= lambda x: x not in rows_wanted)

字符串

对于大文件，一种优化的方法是以 binary 模式打开文件，然后从f.seek结尾开始移动文件指针，检测2个换行符b'\n'（表示各自的最后2行），并将这些行作为io.BinaryIO读入一个字符串。
样品：

import os
import io

fname = 'test.csv'

with open(fname, 'rb') as f:
    head = f.readline() # read the 1st line (head)
    f.seek(-2, os.SEEK_END) # forward file pointer 2 bytes from the end
    n = 2

    while n:
        f.seek(-2, os.SEEK_CUR) # move file pointer backward
        if f.read(1) == b'\n': # catch a linebreak
            n -= 1

    df = pd.read_csv(io.BytesIO(head + f.read()))

字符串

一种尊重文本编码和可能引用的换行符等的方法是使用csv和deque来存储“最后n”行：

import csv
from collections import deque

def row_selector_gen(file, rows, **fmtargs):
    buflen = max(1, 1 - min(rows))
    buflines = deque(maxlen=buflen)
    reader = csv.reader(file, **fmtargs)
    for i, row in enumerate(reader):
        if i in rows:
            yield row
            buflines.append(None)
        else:
            buflines.append(row)
    for i in sorted([i for i in rows if i < 0]):
        try:
            row = buflines[i]
            if row is not None:
                yield row
        except IndexError:
            pass

def row_selector(file, rows, **fmtargs):
    buf = io.StringIO()
    writer = csv.writer(buf, **fmtargs)
    writer.writerows(row_selector_gen(file, rows, **fmtargs))
    buf.seek(0)
    return buf

字符串
范例：

# 1. setup

content = """foo,bar,garçon
1,0.1,"this
and that"
2,0.2,x
3,0.3,y
4,0.4,z
5,0.5,"ha
ha"
"""

with open('test.csv', 'w') as f:
    f.write(content)

型
然后又道：

with open('test.csv', newline='') as f:
    z = pd.read_csv(row_selector(f, [0, 1, -2, -1]))

>>> z
   foo  bar          garçon
0    1  0.1  this\nand that
1    4  0.4               z
2    5  0.5          ha\nha

型

限制：

• 你可能会得到无序的行，例如，如果你在rows中混合了正索引和负索引，或者如果一些负索引引用了一些正索引之前的行。例如，上面的test.csv：

with open('test.csv', newline='') as f:
    z = pd.read_csv(row_selector(f, [0, 3, -4, -1]))

>>> z
   foo  bar  garçon
0    3  0.3       y
1    2  0.2       x
2    5  0.5  ha\nha

型

• 如果您指定了一个早于文件开头的负数行，它们将被默认跳过：

with open('test.csv', newline='') as f:
    z = pd.read_csv(row_selector(f, [0, 3, -14, -1]))

>>> z
   foo  bar  garçon
0    3  0.3       y
1    5  0.5  ha\nha

型