Import Numpy and Pandas:

>>> import numpy as np
>>> import pandas as pd

Create a 5 rows and 3 columns frame with random integers between 0 and 99:

>>> df = pd.DataFrame(np.random.randint(100, size=(5, 3)))
>>> df
    0   1   2
0  35  66  14
1  30   3  13
2  17  69  97
3  99  27   0
4  30  53  64

Add labels to columns:

>>> df.columns = ['a', 'b', 'c']
>>> df
    a   b   c
0  35  66  14
1  30   3  13
2  17  69  97
3  99  27   0
4  30  53  64

Drop all but a and c columns:

>>> df[['a', 'c']]
    a   c
0  35  14
1  30  13
2  17  97
3  99   0
4  30  64

Get a NumPy array of index values:

>>> df.index.values
array([0, 1, 2, 3, 4])

Check if column a is already sorted by comparing initial and value-sorted indexes:

>>> df.a.index.tolist()
[0, 1, 2, 3, 4]
>>> df.a.sort_values().index.tolist()
[2, 4, 1, 0, 3]
>>> df.a.index.tolist() == df.a.sort_values().index.tolist()
False

Make column a the index:

>>> df.set_index('a', inplace=True)
>>> df
    b   c
a
35  66  14
30   3  13
17  69  97
99  27   0
30  53  64

Sort along the index:

>>> df.sort_index(inplace=True)
>>> df
    b   c
a
17  69  97
30  53  64
30   3  13
35  66  14
99  27   0

Deduplicate c data points at the same a index, with the highest c value taking precedence:

>>> df['c'].reset_index().groupby('a').max()
    c
a
17  97
30  64
35  14
99   0

Transform a timeline of arrow objects to Pandas’ internal Timestamp index:

>>> df = pd.DataFrame({'int_ts': pd.Series(np.random.randint(9999999999, size=5))})
>>> df
      int_ts
0  761088975
1  900402905
2  924263705
3  636666598
4  501201802

>>> import arrow
>>> df['dt_arrow'] = df.int_ts.map(arrow.get)
>>> df
      int_ts                   dt_arrow
0  761088975  1994-02-12T21:36:15+00:00
1  900402905  1998-07-14T07:55:05+00:00
2  924263705  1999-04-16T11:55:05+00:00
3  636666598  1990-03-05T19:49:58+00:00
4  501201802  1985-11-18T22:43:22+00:00

>>> from operator import attrgetter
>>> df['dt_index'] = pd.to_datetime(df['dt_arrow'].apply(attrgetter('datetime')), utc=True)
>>> df
      int_ts                   dt_arrow            dt_index
0  761088975  1994-02-12T21:36:15+00:00 1994-02-12 21:36:15
1  900402905  1998-07-14T07:55:05+00:00 1998-07-14 07:55:05
2  924263705  1999-04-16T11:55:05+00:00 1999-04-16 11:55:05
3  636666598  1990-03-05T19:49:58+00:00 1990-03-05 19:49:58
4  501201802  1985-11-18T22:43:22+00:00 1985-11-18 22:43:22

>>> df.set_index('dt_index', inplace=True)
>>> df.sort_index(inplace=True)
>>> df
                        int_ts                   dt_arrow
dt_index
1985-11-18 22:43:22  501201802  1985-11-18T22:43:22+00:00
1990-03-05 19:49:58  636666598  1990-03-05T19:49:58+00:00
1994-02-12 21:36:15  761088975  1994-02-12T21:36:15+00:00
1998-07-14 07:55:05  900402905  1998-07-14T07:55:05+00:00
1999-04-16 11:55:05  924263705  1999-04-16T11:55:05+00:00

Now that we have a properly indexed timeline, we can use built-in Pandas methods. Here is how to compute the maximum value of samples per year :

>>> df['int_ts'].resample('AS')
dt_index
1985-01-01    501201802
1986-01-01          NaN
1987-01-01          NaN
1988-01-01          NaN
1989-01-01          NaN
1990-01-01    636666598
1991-01-01          NaN
1992-01-01          NaN
1993-01-01          NaN
1994-01-01    761088975
1995-01-01          NaN
1996-01-01          NaN
1997-01-01          NaN
1998-01-01    900402905
1999-01-01    924263705
Freq: AS-JAN, Name: int_ts, dtype: float64

Same as above but taking the highest value by shifting decade:

>>> df['int_ts'].resample('10AS', how=max)
dt_index
1985-01-01    761088975
1995-01-01    924263705
Freq: 10AS-JAN, Name: int_ts, dtype: int64

Other resources: