使pandas.read_csv（）在csv文件的开头忽略垃圾？

Question

I've got some junk at the start of my csv file that prevents me selecting the first column of my dataframe by name.

Example:

In[1]: df = pd.read_csv('file:inputdata.csv', usecols=[0], nrows=1)

In[2]: df
Out[2]:
        TAB
0  10-LV_Non

In[3]: df['TAB']
Out[3]: <snip> KeyError: 'TAB'

I found the junk by reading the file with open():

In[4]: with open('inputdata.csv', 'rb') as f:
           print(f.read(7))
Out[4]: b'\xef\xbb\xbfTAB,'

EDIT: '\\xef\\xbb\\xbf' is three bytes of junk. 'TAB' is the name of the first column.

Is there a way to make pandas.read_csv() ignore junks like this (if present) at the start of the csv file?

NB The csv files are exported from a proprietary system, so I can't control their format.

UPDATE: Here's my solution, based on Mike Müller's answer:

with open('inputdata.csv', 'r') as f:
    # Skip past any bytes that aren't text
    while re.match('[a-zA-Z0-9_]', f.read(1)) is None:
        pass
    # Seek back one byte
    f.seek(f.tell()-1)
    # Read the file
    df = pd.read_csv(f, usecols=['TAB'])

Answer 1

It's unclear to me what exactly is the format of the "junk", but there are a number of options to use.

---

pandas.read_csv takes a filepath_or_buffer

filepath_or_buffer : string or file handle / StringIO

It follows that if you open a File object , read past the junk, then pass the File object to read_csv , it should be OK.

---

The skiprows arguments skips rows:

skiprows : list-like or integer, default None

Thus you can possibly skip the junk's row(s).

Answer 2

Something like this could work:

with open('inputdata.csv', 'rb') as f:
    if f.read(7) != b'\xef\xbb\xbfTAB,':
        f.seek(0)
    df = pd.read_csv(f, usecols=[0], nrows=1)

Just read the first seven bytes. If the are good, ie not equal to the bytes you don't want, go back to the beginning of the file with seek(0) , otherwise start reading at position 7 bytes, skipping the offending bytes.