How to get rid of NaN in SFrame python?

I modified the question a bit since you are not using pandas DataFrames.

Because you do not use pandas you should not use your documentation . The documentation of the SFrames indicates that axis is not a valid keyword but you can use columns . Reading your documentation you should be able to do what you want from the following way:

item_info.dropna(columns=None, how='all')

It seems that the error is not thrown dropna but some of the methods you use does not admit that there are columns with some null value.

Starting from an example:

>>> import numpy as np
>>> sf = graphlab.SFrame({'a': [1, None, np.nan],
                          'b': [3, None,   None],
                          'c': [4,    5, np.nan], 
                          'd': [6,    7,      8]})
>>> sf
Data:
+------+------+-----+---+
|  a   |  b   |  c  | d |
+------+------+-----+---+
| 1.0  |  3   | 4.0 | 6 |
| None | None | 5.0 | 7 |
| nan  | None | nan | 8 |
+------+------+-----+---+

dropna delete the entire row or rows depending on the how parameter:

• how = 'all' : Delete only the row if you have todos los valores nulos in all columns specified in columns :

  >>> sf.dropna(columns=None, how='all')   
  Data:
  +------+------+-----+---+
  |  a   |  b   |  c  | d |
  +------+------+-----+---+
  | 1.0  |  3   | 4.0 | 6 |
  | None | None | 5.0 | 7 |
  | nan  | None | nan | 8 |
  +------+------+-----+---+
  
  >>> sf.dropna(columns=['a', 'b'], how='all')
  Data:
  +-----+---+-----+---+
  |  a  | b |  c  | d |
  +-----+---+-----+---+
  | 1.0 | 3 | 4.0 | 6 |
  +-----+---+-----+---+
  

• how = 'any' : Delete a row if al menos un valor nulo exists in some of the columns specified in columns .

  >>> sf.dropna(columns=None, how='any')
  Data:
  +-----+---+-----+---+
  |  a  | b |  c  | d |
  +-----+---+-----+---+
  | 1.0 | 3 | 4.0 | 6 |
  +-----+---+-----+---+
  
  
  >>> sf.dropna(columns=['c', 'd'], how='any')
  Data:
  +------+------+-----+---+
  |  a   |  b   |  c  | d |
  +------+------+-----+---+
  | 1.0  |  3   | 4.0 | 6 |
  | None | None | 5.0 | 7 |
  +------+------+-----+---+
  

As you see, it works with both None and values nan real. If you apply sf.dropna(columns=None, how='all') you will only delete the rows with all their null values but not things like [None, None, 5.0, 7] . Also be careful with null values not recognized as such as 'NaN' especially if you read data from a csv.

If any of the functions you have does not admit that there are null values in the SFrame or you eliminate any row that has at least a null value how='any' or you use fillna to pass those null values to 0 for example.

This depends on your data, its nature and how and what you are evaluating. Deleting a row with a single null value may be losing data and this is not acceptable. In the case of passing the null values to 0 is the same, there are cases in which None can be equivalent to 0 and cases in which this is not true. You must evaluate this and act on consequences, if the problem is the one I expose.

Sframe.fillna has as arguments the column on which to apply the operation and the value by which to replace the nulls:

>>> sf.fillna(column='a', value=0.0)
Data:
+-----+------+-----+---+
|  a  |  b   |  c  | d |
+-----+------+-----+---+
| 1.0 |  3   | 4.0 | 6 |
| 0.0 | None | 5.0 | 7 |
| 0.0 | None | nan | 8 |
+-----+------+-----+---+