python - Finding consecutive segments in a pandas data frame

Question

I have a pandas.DataFrame with measurements taken at consecutive points in time. Along with each measurement the system under observation had a distinct state at each point in time. Hence, the DataFrame also contains a column with the state of the system at each measurement. State changes are much slower than the measurement interval. As a result, the column indicating the states might look like this (index: state):

1:  3
2:  3
3:  3
4:  3
5:  4
6:  4
7:  4
8:  4
9:  1
10: 1
11: 1
12: 1
13: 1

Is there an easy way to retrieve the indices of each segment of consecutively equal states. That means I would like to get something like this:

[[1,2,3,4], [5,6,7,8], [9,10,11,12,13]]

The result might also be in something different than plain lists.

The only solution I could think of so far is manually iterating over the rows, finding segment change points and reconstructing the indices from these change points, but I have the hope that there is an easier solution.

Answer 1

One-liner:

df.reset_index().groupby('A')['index'].apply(np.array)

Code for example:

In [1]: import numpy as np

In [2]: from pandas import *

In [3]: df = DataFrame([3]*4+[4]*4+[1]*4, columns=['A'])
In [4]: df
Out[4]:
    A
0   3
1   3
2   3
3   3
4   4
5   4
6   4
7   4
8   1
9   1
10  1
11  1

In [5]: df.reset_index().groupby('A')['index'].apply(np.array)
Out[5]:
A
1    [8, 9, 10, 11]
3      [0, 1, 2, 3]
4      [4, 5, 6, 7]

You can also directly access the information from the groupby object:

In [1]: grp = df.groupby('A')

In [2]: grp.indices
Out[2]:
{1L: array([ 8,  9, 10, 11], dtype=int64),
 3L: array([0, 1, 2, 3], dtype=int64),
 4L: array([4, 5, 6, 7], dtype=int64)}

In [3]: grp.indices[3]
Out[3]: array([0, 1, 2, 3], dtype=int64)

To address the situation that DSM mentioned you could do something like:

In [1]: df['block'] = (df.A.shift(1) != df.A).astype(int).cumsum()

In [2]: df
Out[2]:
    A  block
0   3      1
1   3      1
2   3      1
3   3      1
4   4      2
5   4      2
6   4      2
7   4      2
8   1      3
9   1      3
10  1      3
11  1      3
12  3      4
13  3      4
14  3      4
15  3      4

Now groupby both columns and apply the lambda function:

In [77]: df.reset_index().groupby(['A','block'])['index'].apply(np.array)
Out[77]:
A  block
1  3          [8, 9, 10, 11]
3  1            [0, 1, 2, 3]
   4        [12, 13, 14, 15]
4  2            [4, 5, 6, 7]