问题
I am trying to load a dataset into pandas and cannot get seem to get past step 1. I am new so please forgive if this is obvious, I have searched previous topics and not found an answer. The data is mostly in Chinese characters, which may be the issue.
The .csv is very large, and can be found here: http://weiboscope.jmsc.hku.hk/datazip/ I am trying on week 1.
In my code below, I identify 3 types of decoding I attempted, including an attempt to see what encoding was used
import pandas
import chardet
import os
#this is what I tried to start
data = pandas.read_csv('week1.csv', encoding="utf-8")
#spits out error: UnicodeDecodeError: 'utf-8' codec can't decode byte 0x9a in position 69: invalid start byte
#Code to check encoding -- this spits out ascii
bytes = min(32, os.path.getsize('week1.csv'))
raw = open('week1.csv', 'rb').read(bytes)
chardet.detect(raw)
#so i tried this! it also fails, which isn't that surprising since i don't know how you'd do chinese chars in ascii anyway
data = pandas.read_csv('week1.csv', encoding="ascii")
#spits out error: UnicodeDecodeError: 'ascii' codec can't decode byte 0xe6 in position 0: ordinal not in range(128)
#for god knows what reason this allows me to load data into pandas, but definitely not correct encoding because when I print out first 5 lines its gibberish instead of Chinese chars
data = pandas.read_csv('week1.csv', encoding="latin1")
Any help would be greatly appreciated!
EDIT: The answer provided by @Kristof does in fact work, as does the program a colleague of mine put together yesterday:
import csv
import pandas as pd
def clean_weiboscope(file, nrows=0):
res = []
with open(file, 'r', encoding='utf-8', errors='ignore') as f:
reader = csv.reader(f)
for i, row in enumerate(f):
row = row.replace('\n', '')
if nrows > 0 and i > nrows:
break
if i == 0:
headers = row.split(',')
else:
res.append(tuple(row.split(',')))
df = pd.DataFrame(res)
return df
my_df = clean_weiboscope('week1.csv', nrows=0)
I also wanted to add for future searchers that this is the Weiboscope open data for 2012.
回答1:
It seems that there's something very wrong with the input file. There are encoding errors throughout.
One thing you could do, is to read the CSV file as a binary, decode the binary string and replace the erroneous characters.
Example (source for the chunk-reading code):
in_filename = 'week1.csv'
out_filename = 'repaired.csv'
from functools import partial
chunksize = 100*1024*1024 # read 100MB at a time
# Decode with UTF-8 and replace errors with "?"
with open(in_filename, 'rb') as in_file:
with open(out_filename, 'w') as out_file:
for byte_fragment in iter(partial(in_file.read, chunksize), b''):
out_file.write(byte_fragment.decode(encoding='utf_8', errors='replace'))
# Now read the repaired file into a dataframe
import pandas as pd
df = pd.read_csv(out_filename)
df.shape
>> (4790108, 11)
df.head()
来源:https://stackoverflow.com/questions/38728366/pandas-cannot-load-data-csv-encoding-mystery