Interfacing with the Pandas Package

The pandas package is a package for high performance data analysis of table-like structures that is complementary to the Table class in astropy.

In order to exchange data between the Table class and the pandas DataFrame class (the main data structure in pandas), the Table class includes two methods, to_pandas() and from_pandas().


To demonstrate, we can create a minimal table:

>>> from astropy.table import Table
>>> t = Table()
>>> t['a'] = [1, 2, 3, 4]
>>> t['b'] = ['a', 'b', 'c', 'd']

Which we can then convert to a pandas DataFrame:

>>> df = t.to_pandas()
>>> df
   a  b
0  1  a
1  2  b
2  3  c
3  4  d
>>> type(df)
<class 'pandas.core.frame.DataFrame'>

It is also possible to create a table from a DataFrame:

>>> t2 = Table.from_pandas(df)
>>> t2
<Table length=4>
  a      b
int64 string8
----- -------
    1       a
    2       b
    3       c
    4       d

The conversions to and from pandas are subject to the following caveats:

  • The pandas DataFrame structure does not support multidimensional columns, so Table objects with multidimensional columns cannot be converted to DataFrame.

  • Masked tables can be converted, but DataFrame uses numpy.nan to indicate masked values, so all numerical columns (integer or float) are converted to numpy.float columns in DataFrame, and string columns with missing values are converted to object columns with numpy.nan values to indicate missing values. For numerical columns, the conversion therefore does not necessarily round-trip if converting back to an astropy table, because the distinction between numpy.nan and masked values is lost, and the different integer columns (for example) will be converted to floating- point.

  • Tables with mixin columns can currently not be converted, but this may be implemented in the future.