Pandas read_csv for a no quote file

Question

I'm trying to read a file that doesn't have any quotes, which is causing inconsistent number of row lengths

Data looks as follows:

col_a, col_b
abc, inc., 5
xyz corb, 10

Since there are no quotes around "abc, inc.", this is causing the first row to get split into 3 values, but it should actually be just 2 values.

This column is not necessarily in the first position, and that there can be another bad column like this. The data has around 250 columns.

I'm reading this using pd.read_csv, how can this be resolved?

Thanks!

Answer 1

EDIT (after updated question with an additional column):

Solution 1:

You can create a function with the bad column as a parameter and use split and concat to correct the dataframe depending on that bad column. Please note that the bad_col parameter in my function is the column number, where we start counting at 1, rather than 0 (e.g. 1, 2, 3, etc. instead of 0, 1, 2, etc.):

import pandas as pd
import numpy as np
from io import StringIO
data = StringIO('''
col, col_a, col_b
000, abc, inc., 5
111, xyz corb, 10
''')
df = pd.read_csv(data, sep="|")

def fix_csv(df, bad_col):
    cols = df.columns.str.split(', ')[0]
    x = len(cols) - bad_col
    tmp = df.iloc[:,0].str.split(', ', expand=True, n=x)
    df = pd.concat([tmp.iloc[:,0],
                    tmp.iloc[:,-1].str.rsplit(', ', expand=True, n=x)],
                    axis=1)
    df.columns = cols
    return df


fix_csv(df, bad_col=2)

---

Solution 2 (this is if you have issues in multiple columns and you need to use more brute force):

It sounds like there is a possibility that you there could be multiple columns affected from the comments as you mentioned only 1 "so far".

As such, this might be a little bit of a project to clean up the data. The following code can give you an idea how to do that. The bottom-line is that you can create two different dataframes: 1) The first dataframe has the minimum number of commas (i.e. they should be the rows without any issues). 2) The other dataframe will be the dataframe with all of the issues. I've shown how you can clean the data to get to the correct number of columns and then change the data back and concat the two dataframes.

import pandas as pd
import numpy as np
from io import StringIO
data = StringIO('''
col, col_a, col_b
000, abc, inc., 5
111, xyz corb, 10
''')
df = pd.read_csv(data, sep="|")
cols = df.columns.str.split(', ')[0]
s = df.iloc[:,0].str.count(',')
df1 = df.copy()[s.eq(s.min())]
df1 = df1.iloc[:,0].str.split(', ', expand=True)
df1.columns = cols
df2 = df.copy()[s.gt(s.min())]
#inspect this dataframe manually to see how many rows affected, which columns, etc.
#cleanup df2 with some .replace so all equal commas
original = [', inc.', ', corp.']
temp = [' inc.', ' corp.']
df2.iloc[:,0] = df2.iloc[:,0].replace(original, temp, regex=True)
df2 = df2.iloc[:,0].str.split(', ', expand=True)
df2.columns = cols
#cleanup df2 by changing back to original values
df2['col_a'] = df2['col_a'].replace(temp, original, regex=True) # you can do this with other columns as well
df3 = pd.concat([df1, df2]).sort_index()
df3
Out[1]: 
   col      col_a col_b
0  000  abc, inc.     5
1  111   xyz corb    10

---

Solution 3: Previous Solution (for original question when problem was only in first column - for reference)

1. You can read in with sep="|" as that | character is not in your .csv, so it reads all of the data into one column.
2. The main assumption to my solution is that the problematic column is only the first column. I use rsplit(', ') and limit the number of splits to the total number of columns minus 1 (with the example data, this is 2-1=1). Hopefully, this solves with your actual data or at least gives you some idea. If your data is separated by , instead of , , please note whether or not to adjust my splits as well.

---

import pandas as pd
import numpy as np
from io import StringIO
data = StringIO('''
col_a, col_b
abc, inc., 5
xyz corb, 10
''')
df = pd.read_csv(data, sep="|")
cols = df.columns.str.split(', ')[0]
x = len(cols) - 1
df = df.iloc[:,0].str.rsplit(', ', expand=True, n=x)
df.columns = cols
df
Out[1]: 
       col_a col_b
0  abc, inc.     5
1   xyz corb    10

Answer 2

EDIT: Thanks to tdelaney comment below: see if this works pd.read_csv('foo.csv',delimiter=",(?!( [\w\d]*).,)").dropna(axis=1)

OLD: using delimiter as ",(?!.*,)" in read_csv seems to be solving this for me

Answer 3

Its not a CSV but since there is only one column with the errant commas you can process with the csv module and fix the slice that holds too many column values. When a row has too many cells, assume they are the ones from the unescaped comma.

import pandas as pd
import csv

def split_badrows(fileobj, bad_col, total_cols):
    """Iterate rows, colapsing extra columns at bad_col"""
    for row in csv.reader(fileobj):
        row = [cell.strip() for cell in row]
        extras = len(row) - total_cols
        if extras > 0:
            # colapse slice at troubled column into single value
            extras += 1 # python slice doesn't include right endpoint
            row[bad_col] = ", ".join(row[bad_col:bad_col+extras])
            del row[bad_col+1:bad_col+extras]
        yield row

def df_from_badtext(fileobj, bad_col):
    """Make pandas.DataFrame from badly formatted text"""
    columns = [cell.strip() for cell in next(fileobj).split(",")]
    total_cols = len(columns)
    return pd.DataFrame(split_badrows(fileobj, bad_col, total_cols),
            columns=columns)

# test

open("testme.txt", "w").write("""col_a, col_b
abc, inc., 5
xyz corb, 10""")

df = df_from_badtext(open("testme.txt"), bad_col=0)
print(df)

Answer 4

Data split to list then transform to dataframe.

csv = '''col_a, col_b
abc, inc., 5
xyz corb, 10'''+'\n'
import re
import pandas as pd
reArr = re.findall('(.*),([^,]+)\n',csv)
df=pd.DataFrame(reArr[1:],columns=reArr[0])
print(df)

	col_a	col_b
0	abc, inc.	5
1	xyz corb	10