小开

Assuming no header in the CSV file:

import pandas
import random


n = 1000000 #number of records in file
s = 10000 #desired sample size
filename = "data.txt"
skip = sorted(random.sample(range(n),n-s))
df = pandas.read_csv(filename, skiprows=skip)

would be better if read_csv had a keeprows, or if skiprows took a callback func instead of a list.

With header and unknown file length:

import pandas
import random


filename = "data.txt"
n = sum(1 for line in open(filename)) - 1 #number of records in file (excludes header)
s = 10000 #desired sample size
skip = sorted(random.sample(range(1,n+1),n-s)) #the 0-indexed header will not be included in the skip list
df = pandas.read_csv(filename, skiprows=skip)

小开

class magic_checker:
def __init__(self,target_count):
self.target = target_count
self.count = 0
def __eq__(self,x):
self.count += 1
return self.count >= self.target


min_target=100000
max_target = min_target*2
nlines = randint(100,1000)
seek_target = randint(min_target,max_target)
with open("big.csv") as f:
f.seek(seek_target)
f.readline() #discard this line
rand_lines = list(iter(lambda:f.readline(),magic_checker(nlines)))


#do something to process the lines you got returned .. perhaps just a split
print rand_lines
print rand_lines[0].split(",")

something like that should work I think

小开

No pandas!

import random
from os import fstat
from sys import exit


f = open('/usr/share/dict/words')


# Number of lines to be read
lines_to_read = 100


# Minimum and maximum bytes that will be randomly skipped
min_bytes_to_skip = 10000
max_bytes_to_skip = 1000000


def is_EOF():
return f.tell() >= fstat(f.fileno()).st_size


# To accumulate the read lines
sampled_lines = []


for n in xrange(lines_to_read):
bytes_to_skip = random.randint(min_bytes_to_skip, max_bytes_to_skip)
f.seek(bytes_to_skip, 1)
# After skipping "bytes_to_skip" bytes, we can stop in the middle of a line
# Skip current entire line
f.readline()
if not is_EOF():
sampled_lines.append(f.readline())
else:
# Go to the begginig of the file ...
f.seek(0, 0)
# ... and skip lines again
f.seek(bytes_to_skip, 1)
# If it has reached the EOF again
if is_EOF():
print "You have skipped more lines than your file has"
print "Reduce the values of:"
print "   min_bytes_to_skip"
print "   max_bytes_to_skip"
exit(1)
else:
f.readline()
sampled_lines.append(f.readline())


print sampled_lines

You'll end up with a sampled_lines list. What kind of statistics do you mean?

小开

The following code reads first the header, and then a random sample on the other lines:

import pandas as pd
import numpy as np


filename = 'hugedatafile.csv'
nlinesfile = 10000000
nlinesrandomsample = 10000
lines2skip = np.random.choice(np.arange(1,nlinesfile+1), (nlinesfile-nlinesrandomsample), replace=False)
df = pd.read_csv(filename, skiprows=lines2skip)

小开

Here is an algorithm that doesn't require counting the number of lines in the file beforehand, so you only need to read the file once.

Say you want m samples. First, the algorithm keeps the first m samples. When it sees the i-th sample (i > m), with probability m/i, the algorithm uses the sample to randomly replace an already selected sample.

By doing so, for any i > m, we always have a subset of m samples randomly selected from the first i samples.

See code below:

import random


n_samples = 10
samples = []


for i, line in enumerate(f):
if i < n_samples:
samples.append(line)
elif random.random() < n_samples * 1. / (i+1):
samples[random.randint(0, n_samples-1)] = line

小开

This is not in Pandas, but it achieves the same result much faster through bash, while not reading the entire file into memory:

shuf -n 100000 data/original.tsv > data/sample.tsv

The shuf command will shuffle the input and the and the -n argument indicates how many lines we want in the output.

Relevant question: https://unix.stackexchange.com/q/108581

Benchmark on a 7M lines csv available here (2008):

Solution 1: Approximate Percentage

If you can specify what percent of lines you want, rather than how many lines, you don't even need to get the file size and you just need to read through the file once. Assuming a header on the first row:

import pandas as pd
import random
p = 0.01  # 1% of the lines
# keep the header, then take only 1% of lines
# if random from [0,1] interval is greater than 0.01 the row will be skipped
df = pd.read_csv(
filename,
header=0,
skiprows=lambda i: i>0 and random.random() > p
)

As pointed out in the comments, this only gives approximately the right number of lines, but I think it satisfies the desired usecase.

Solution 2: Every Nth line

This isn't actually a random sample, but depending on how your input is sorted and what you're trying to achieve, this may meet your needs.

n = 100  # every 100th line = 1% of the lines
df = pd.read_csv(filename, header=0, skiprows=lambda i: i % n != 0)

Solution 3: Reservoir Sampling

(Added July 2021)

Reservoir sampling is an elegant algorithm for selecting k items randomly from a stream whose length is unknown, but that you only see once.

The big advantage is that you can use this without having the full dataset on disk, and that it gives you an exactly-sized sample without knowing the full dataset size. The disadvantage is that I don't see a way to implement it in pure pandas, I think you need to drop into python to read the file and then construct the dataframe afterwards. So you may lose some functionality from read_csv or need to reimplement it, since we're not using pandas to actually read the file.

Taking an implementation of the algorithm from Oscar Benjamin here:

from math import exp, log, floor
from random import random, randrange
from itertools import islice
from io import StringIO


def reservoir_sample(iterable, k=1):
"""Select k items uniformly from iterable.


Returns the whole population if there are k or fewer items


from https://bugs.python.org/issue41311#msg373733
"""
iterator = iter(iterable)
values = list(islice(iterator, k))


W = exp(log(random())/k)
while True:
# skip is geometrically distributed
skip = floor( log(random())/log(1-W) )
selection = list(islice(iterator, skip, skip+1))
if selection:
values[randrange(k)] = selection[0]
W *= exp(log(random())/k)
else:
return values


def sample_file(filepath, k):
with open(filepath, 'r') as f:
header = next(f)
result = [header] + sample_iter(f, k)
df = pd.read_csv(StringIO(''.join(result)))

The reservoir_sample function returns a list of strings, each of which is a single row, so we just need to turn it into a dataframe at the end. This assumes there is exactly one header row, I haven't thought about how to extend it to other situations.

I tested this locally and it is much faster than the other two solutions. Using a 550 MB csv (January 2020 "Yellow Taxi Trip Records" from the NYC TLC), solution 3 runs in about 1 second, while the other two take ~3-4 seconds.

In my test this is even slightly (~10-20%) faster than @Bar's answer using shuf, which surprises me.

小开

use subsample

pip install subsample
subsample -n 1000 file.csv > file_1000_sample.csv

小开

You can also create a sample with the 10000 records before bringing it into the Python environment.

Using Git Bash (Windows 10) I just ran the following command to produce the sample

shuf -n 10000 BIGFILE.csv > SAMPLEFILE.csv

To note: If your CSV has headers this is not the best solution.

小开

For example, you have the loan.csv, you can use this script to easily load the specified number of random items.

data = pd.read_csv('loan.csv').sample(10000, random_state=44)

小开

Let's say that you want to load a 20% sample of the dataset:

    import pandas as pd
df = pd.read_csv(filepath).sample(frac = 0.20)

小开

TL;DR

If you know the size of the sample you want, but not the size of the input file, you can efficiently load a random sample out of it with the following pandas code:

import pandas as pd
import numpy as np


filename = "data.csv"
sample_size = 10000
batch_size = 200


rng = np.random.default_rng()


sample_reader = pd.read_csv(filename, dtype=str, chunksize=batch_size)


sample = sample_reader.get_chunk(sample_size)


for chunk in sample_reader:
chunk.index = rng.integers(sample_size, size=len(chunk))
sample.loc[chunk.index] = chunk

Explanation

It's not always trivial to know the size of the input CSV file.

If there are embedded line breaks, tools like wc or shuf will give you the wrong answer or just make a mess out of your data.

So, based on desktable's answer, we can treat the first sample_size lines of the file as the initial sample and then, for each subsequent line in the file, randomly replace a line in the initial sample.

To do that efficiently, we load the CSV file using a TextFileReader by passing the chunksize= parameter:

sample_reader = pd.read_csv(filename, dtype=str, chunksize=batch_size)

First, we get the initial sample:

sample = sample_reader.get_chunk(sample_size)

Then, we iterate over the remaining chunks of the file, replacing the index of each chunk with a sequence of random integers as long as size of the chunk, but where each integer is in the range of the index of the initial sample (which happens to be the same as range(sample_size)):

for chunk in sample_reader:
chunk.index = rng.integers(sample_size, size=len(chunk))

And use this reindexed chunk to replace (some of the) lines in the sample:

sample.loc[chunk.index] = chunk

After the for loop, you'll have a dataframe at most sample_size rows long, but with random lines selected from the big CSV file.

To make the loop more efficient, you can make batch_size as large as your memory allows (and yes, even larger than sample_size if you can).

Notice that, while creating the new chunk index with np.random.default_rng().integers(), we use len(chunk) as the new chunk index size instead of simply batch_size because the last chunk in the loop could be smaller.

On the other hand, we use sample_size instead of len(sample) as the "range" of the random integers, even though there could be less lines in the file than sample_size. This is because there won't be any chunks left to loop over in this case so that will never be a problem.

小开

read the data file

import pandas as pd
df = pd.read_csv('data.csv', 'r')

First check the shape of df

df.shape()

create the small sample of 1000 raws from df

sample_data = df.sample(n=1000, replace='False')

#check the shape of sample_data

sample_data.shape()

将一个小的随机样本从一个大的 CSV 文件读入 Python 数据帧