43

Is there a way to preserve the order of the columns in a csv file when read and the write with Python Pandas? For example, in this code

import pandas as pd

data = pd.read_csv(filename)
data.to_csv(filename)

the output files might be different because the columns are not preserved.

Hernan
  • 5,199
  • 5
  • 46
  • 79
  • Can you provide an example of your csv? – waitingkuo Mar 27 '13 at 08:09
  • 4
    wishing OP had added a "when the column names are not known in advance" qualifier to this question. All the answers posted here assume that all the columns are already known, even though OP never said so. – Nikhil VJ Jul 05 '18 at 04:44

3 Answers3

36

There appears to be a bug in the current version of Pandas ('0.11.0'), which means that Matti John's answer will not work. If you specify columns for writing to file, they are written in alphabetical order, but simply relabelled according to the list in cols. For example, this code:

import pandas
dfdict={}
dfdict["a"]=[1,2,3,4]
dfdict["b"]=[5,6,7,8]
dfdict["c"]=[9,10,11,12]
df=pandas.DataFrame(dfdict)
df.to_csv("dfTest.txt","\t",header=True,cols=["b","a","c"])

results in this (incorrect) output:

    b   a   c
0   1   5   9
1   2   6   10
2   3   7   11
3   4   8   12

You can check which version of pandas you have installed by executing:

pandas.version.version

Documentation for to_csv is here

Actually, it seems that this is a known bug and will be fixed in an upcoming release (0.11.1):

https://github.com/pydata/pandas/issues/3489

UPDATE: There still hasn't been a new release of pandas, but there is a workaround described here, which doesn't require using a different version of pandas:

github.com/pydata/pandas/issues/3454

So changing the last line in the block of code above to the following will work correctly: 

df.to_csv("dfTest.txt","\t",header=True,cols=["b","a","c"], engine='python')

UPDATE it seems that the argument "cols" has been renamed to "columns" and that the argument "engine" is deprecated (no longer available) in recent versions of pandas. Also, this bug is fixed in version 0.19.0.

CnrL
  • 2,381
  • 18
  • 26
24

The column order should generally be preserved when reading and then writing a csv file like that, but if for some reason they are not in the order you want you can use the columns keyword argument in to_csv.

For example, if you have a csv with columns a, b, c, d:

data = pd.read_csv(filename)
data.to_csv(filename, columns=['a', 'b', 'c', 'd'])
ayhan
  • 57,258
  • 12
  • 148
  • 173
Matti John
  • 15,554
  • 7
  • 35
  • 38
6

Another workaround is to do this:

import pandas as pd
data = pd.read_csv(filename)
data2 = df[['A','B','C']]  #put 'A' 'B' 'C' in the desired order
data2.to_csv(filename)
Maximilian Kohl
  • 612
  • 1
  • 9
  • 21
Lawrence Chernin
  • 133
  • 2
  • 11
  • This was the only solution that worked for me. You could reduce a line of code by reordering and creating the CSV all in one step. – Mtap1 Sep 15 '16 at 16:59