熊猫中DataFrame的矩阵（位置）索引

Question

Say I have the following dataframe:

tmp = np.random.randn(10,4)
df = pd.DataFrame(tmp, index=pd.date_range('1/1/2012', periods=tmp.shape[0]), 
                 columns=['A', 'B', 'C', 'D'])

> b
                   A         B         C         D
2012-01-01  0.471846  1.130041 -0.614117  0.882738
2012-01-02 -1.431566  0.680617 -0.615331  0.288740
2012-01-03  0.398567 -0.115388 -0.869855 -1.273666
2012-01-04  0.379501  0.192329 -1.942184  0.694004
2012-01-05  1.306329 -0.803856  0.417033 -0.655907
2012-01-06 -0.599877  0.696549 -0.252789  1.367977
2012-01-07 -1.618916  0.216571 -0.499880  0.386853
2012-01-08  0.415002  0.139775  0.251842  0.021379
2012-01-09  2.536787  0.737672 -0.740485 -0.890189
2012-01-10 -1.553530 -0.100950 -0.237478 -0.295612

How can I do:

1. Positional indexing of specific rows/columns? (and get the corresponding sub- dataframe )
2. Positional indexing of ranges of rows/columns? (and get the corresponding sub- dataframe )

For single-entry matricial indexing:

For example, say I want to index the sub-dataframe in location [1,2] (in numpy "matricial" notation). The output should be:

                   C
2012-01-02 -0.615331

I tried the following three methods, but none of them worked::

df[1,2]
df[1][2]
df.take([1])[2]

The only methods that work seem to be:

df.ix[1,2]
df.irow(1)[2]

but:

• Using .ix for positional indexing is dangerous, since it would default to label indexing if my indices were integers (as opposed to dates as in the case above). See more on this here: Start:stop slicing inconsistencies between numpy and Pandas? .

• Using irow is cumbersome, since it requires switching from () notation to [] notation ( irow returns a Series object)

For range matricial indexing:

For example, say I want to index elements in locations [1:3,2:3] in (numpy matricial notation). The output should be:

                   B
2012-01-02 -0.615331  
2012-01-03 -0.869855 

Note that I am excluding the stop indices (ie I am sticking to the numpy notation).

Any thoughts?

Answer 1

经常会要求使用此功能， https://github.com/pydata/pandas/pull/2922如果您想对其进行测试，可以将其从分支中拉出

Answer 2

Here is a workaround (until the feature request @Jeff mentioned gets committed):

In [178]: df = pd.DataFrame(tmp, index=pd.date_range('2012-1-1', periods=tmp.shape[0]), columns='A B C D'.split())

In [179]: df.ix[df.index[1], df.columns[2]]
Out[179]: -0.3021434106214243

In [180]: df.ix[df.index[1:3], df.columns[2:3]]
Out[180]: 
                   C
2012-01-02 -0.302143
2012-01-03 -1.430387

This shows the syntax works the same way even with shuffled integer indices:

In [206]: df2 = df.reset_index(drop=True)

In [207]: index = range(10)

In [208]: import random

In [209]: random.shuffle(index)

In [210]: df2.index = index

In [212]: df2.ix[df2.index[1], df2.columns[2]]
Out[212]: -0.3021434106214243

In [213]: df2.ix[df2.index[1:3], df2.columns[2:3]]
Out[213]: 
          C
7 -0.302143
2 -1.430387

Answer 3

from the pandas documentation:

Pandas provides a suite of methods in order to get purely integer based indexing. The semantics follow closely python and numpy slicing. These are 0-based indexing. When slicing, the start bounds is included, while the upper bound is excluded. Trying to use a non-integer, even a valid label will raise a IndexError.

The .iloc attribute is the primary access method. The following are valid inputs:

An integer eg 5 A list or array of integers [4, 3, 0] A slice object with ints 1:7