ipython pandas TypeError: read_csv() got an unexpe

问题:

While trying the ipython.org notebook, "INTRODUCTION TO PYTHON FOR DATA MINING"

The following code:

data = pd.read_csv("http://archive.ics.uci.edu/ml/machine-learning-databases/auto-mpg/auto-mpg.data-original",
               delim_whitespace = True, header=None,
               names = ['mpg', 'cylinders', 'displacement', 'horsepower', 'weight', 'acceleration',
                        'model', 'origin', 'car_name'])

yields the following error:

 TypeError: read_csv() got an unexpected keyword argument 'delim-whitespace'

Unfortunately the dataset file itself is not really csv, and I don't know why they used read_csv() to get its data.

The data looks like this line:

 14.0   8.   454.0      220.0      4354.       9.0   70.  1.    "chevrolet impala"

The environment is python/2.7 on Debian stable w/ ipython 0.13. After searching here, I realize it's mostly likely a version problem, as the argument 'delim-whitespace' maybe in a later version of the pandas library, than the one available to the APT package manager.

I tried several workarounds, without success.

• First, I tried to upgrade pandas, by building from latest source, but i found i would end up with a cascade of other builds of dependencies whose versions need upgrading and could end up breaking the environment. E.g., I had to install Cython, then it reported it was again a version too old on the APT package manager, so I would have to rebuild Cython, + other libs/modules and so on.

• Then after looking at the API a bit, I tried using other arguments: using delimiter = ' ' in the call to read_csv() caused it to break up the strings inside quotes into several columns,

  ValueError: Expecting 9 columns, got 13 in row 0
  

• I tried using the read_csv() argument quotechar='"' , as documented in the API but again it was not recognized (unexpected keyword argument)

• Finally I tried using a different way to load the file,

  data = DataFrame()
  
  data.from_csv(url)
  

  I got,

  Out[18]: 
  <class 'pandas.core.frame.DataFrame'>
  Index: 405 entries, 15.0   8.   350.0      165.0      3693.      11.5   70.  1."buick skylark 320" to 31.0   4.   119.0      82.00      2720.      19.4   82.  1.   "chevy s-10"
  Empty DataFrame
  
  In [19]: print(data.shape)
  (0, 9)
  

• alternatively, w/ sep argument to from_csv(),

  In [20]: data.from_csv(url,sep=' ')
  

  yields the error,

  ValueError: Expecting 31 columns, got 35 in row 1
  In [21]: print(data.shape)
  (0, 9)
  

• Also alternatively, with the same negative result:

  In [32]: data = DataFrame( columns = ['mpg', 'cylinders', 'displacement', 'horsepower', 'weight', 'acceleration','model', 'origin', 'car_name'])
  
  In [33]: data.from_csv(url,sep=', \t')Out[33]: 
  <class 'pandas.core.frame.DataFrame'>
  Index: 405 entries, 15.0   8.   350.0      165.0      3693.      11.5   70.  1."buick skylark 320" to 31.0   4.   119.0      82.00      2720.      19.4   82.  1.   "chevy s-10"
  Empty DataFrame
  
  In [34]: data.head()
  Out[34]: 
  Empty DataFrame
  

• I tried using ipython3 instead, but it cannot find/load matplotlib as there is not matplotlib for python3 for my system.

Any help with this problem would be greatly appreciated.

回答1:

Oddly, the delim_whitespace parameter appears in the Pandas documentation in the method summary but not the parameters list. Try replacing it with delimiter = r'\s+', which is equivalent to what I assume the authors meant.

CSV does refer to comma-separated values, but it's often used to refer to general delimited-text formats. TSV (tab-separated values) is another variant; in this case it's basically whitespace-separated values.

回答2:

Your code uses delim_whitespace but the error message says delim-whitespace. The former exists, the latter does not.

If the data file contains

 14.0   8.   454.0      220.0      4354.       9.0   70.  1.    "chevrolet impala"

and you define data with

data = pd.read_csv('data', delim_whitespace = True, header=None, names = ['mpg', 'cylinders', 'displacement', 'horsepower', 'weight', 'acceleration', 'model', 'origin', 'car_name'])

then the DataFrame does get parsed successfully:

   mpg  cylinders  displacement  horsepower  weight  acceleration  model  \
0   14          8           454         220    4354             9     70   

   origin          car_name  
0       1  chevrolet impala  

So you just have change the hyphen to an underscore.

---

Note that when you specify delim_whitespace=True, the pure Python parser is used. In this case I don't think that is necessary. Using delimiter=r'\s+' as Steve Howard suggests would probably perform better. (The source code says, "The C engine is faster while the python engine is currently more feature-complete", but I think the only feature that the python engine has that the C engine does not is skipfooter.)