52

I have a very simple csv, with the following data, compressed inside the tar.gz file. I need to read that in dataframe using pandas.read_csv. 

   A  B
0  1  4
1  2  5
2  3  6

import pandas as pd
pd.read_csv("sample.tar.gz",compression='gzip')

However, I am getting error:

CParserError: Error tokenizing data. C error: Expected 1 fields in line 440, saw 2

Following are the set of read_csv commands and the different errors I get with them:

pd.read_csv("sample.tar.gz",compression='gzip',  engine='python')
Error: line contains NULL byte

pd.read_csv("sample.tar.gz",compression='gzip', header=0)
CParserError: Error tokenizing data. C error: Expected 1 fields in line 440, saw 2

pd.read_csv("sample.tar.gz",compression='gzip', header=0, sep=" ")
CParserError: Error tokenizing data. C error: Expected 2 fields in line 94, saw 14    

pd.read_csv("sample.tar.gz",compression='gzip', header=0, sep=" ", engine='python')
Error: line contains NULL byte

What's going wrong here? How can I fix this?

Geet
  • 2,515
  • 2
  • 19
  • 42

2 Answers2

84
df = pd.read_csv('sample.tar.gz', compression='gzip', header=0, sep=' ', quotechar='"', error_bad_lines=False)

Note: error_bad_lines=False will ignore the offending rows.

Marlon Abeykoon
  • 11,927
  • 4
  • 54
  • 75
  • Thanks, Marlon. What's ".dat" in 3rd line, here? – Geet Sep 01 '16 at 06:34
  • When I try that, it says, KeyError: "filename 'sample.dat' not found" – Geet Sep 01 '16 at 06:41
  • @Geet and also tell me your pandas version. This should work for 0.18.1 – Marlon Abeykoon Sep 01 '16 at 06:45
  • My pandas version is 0.18.1. The updated code give me "CParserError: Error tokenizing data. C error: Expected 1 fields in line 440, saw 2" error – Geet Sep 01 '16 at 06:53
  • 2
    This worked for me for a sample csv file. your link let me download 40GB. don't you have a sample of it for me to test? – Marlon Abeykoon Sep 01 '16 at 06:58
  • Check the code I set `error_bad_lines=False` and just now noticed your sep is an empty string. Can you try again with updated answer – Marlon Abeykoon Sep 01 '16 at 07:02
  • Did it actually read all the data correctly? Using `error_bad_lines` will just cause it to skip lines with errors, so the result may be missing some rows. Those errors might indicate an actual error in the data file. – BrenBarn Sep 01 '16 at 07:20
  • @BrenBarn: You are right, I see too many lines being skipped with that option. Any solution for that? – Geet Sep 01 '16 at 07:23
  • @Geet: I suspect there is some corruption in the file, but it could be hard to pinpoint it if the file is 40G. – BrenBarn Sep 01 '16 at 07:27
  • I have tested with 2-3 much smaller tar.gz files, but still facing the same issue. – Geet Sep 01 '16 at 07:30
  • @Geet: It would be good if you could supply a reasonably-sized sample file (like a few KB) to test with. – BrenBarn Sep 05 '16 at 22:52
  • @BrenBarn: I do have a 1-2 KB file, but how can I supply it here? – Geet Sep 06 '16 at 23:54
  • @Geet: You'll have to upload it somewhere and provide a link, as you did with your original file. – BrenBarn Sep 07 '16 at 02:48
  • For pandas 2.0 it will be like: ```df = pd.read_csv(r"path_to_csv.gz_file",compression="gzip",header=0,sep=",",quotechar='"',on_bad_lines="skip") ``` – Arun Apr 12 '23 at 10:06
14

You can use the tarfile module to read a particular file from the tar.gz archive (as discussed in this resolved issue). If there is only one file in the archive, then you can do this:

import tarfile
import pandas as pd
with tarfile.open("sample.tar.gz", "r:*") as tar:
    csv_path = tar.getnames()[0]
    df = pd.read_csv(tar.extractfile(csv_path), header=0, sep=" ")

The read mode r:* handles the gz extension (or other kinds of compression) appropriately. If there are multiple files in the zipped tar file, then you could do something like csv_path = list(n for n in tar.getnames() if n.endswith('.csv'))[-1] line to get the last csv file in the archived folder.

teichert
  • 3,963
  • 1
  • 31
  • 37