Building new dataframe from series - python-2.7

Question

Taking another crack at an older question of mine since I still do not understand how to properly do what I want.
I have data stored in a dataframe and need to extract averaged chunks of it to use later. My index is datetime values, but this is not terribly important. Unfortunately, I cannot do a simple df.resample() operation, since the data I need to extract is not regularly spaced. Example:
import pandas as pd
from numpy import *
# Build example dataframe
df = pd.DataFrame(data=random.rand(10,3),index=None,columns=list('ABC'))
# Build dummy dataframe to store averaged data from "df"
dummy = pd.DataFrame(columns=df.columns)
# Perform averaging of "df"
for r in xrange(1,10,2):
ave = df.ix[r-1:r+1].mean()
# Store averaged data in dummy dataframe
# Here is where I hit my problem, since ave is a Series
dummy = dummy.append(ave)
I cannot append a series to dataframe.
I can work around by converting ave to a dictionary, then appending to dummy:
for r in xrange(1,10,2):
ave = df.ix[r-1:r+1].mean().to_dict()
ave = pd.DataFrame(ave,index=[r])
dummy = dummy.append(ave)
First: does my overall goal make sense?
Second: Is there a better way to achieve this? Converting to dictionary, then dataframe, then appending seems kludgey, but it is the best I have.
Begin Edit
unutbu raised a good point. As written, rolling_mean() will work. But, I am interested only in very few rows of data, everything else is considered garbage.
# Now creating larger dataframe for illustration
df = pd.DataFrame(data=random.rand(10000,3),index=None,columns=list('ABC'))
# Now, most of the data are not averaged
for r in xrange(1,10000,50):
ave = df.ix[r-1:r+1].mean().to_dict()
ave = pd.DataFrame(ave,index=[r])
The main problem I have with my examples is showing the irregularity with which the averaging is done. The averaging is event driven (i.e. if something happened at 2013-01-01 14:23 then average the data about 2013-01-01 14:23 +/- 2.5min.
Unfortunately, the data timestamps are also highly irregular, which makes rolling_mean() ineffective in this case. So I have irregular events determining when I should average my irregularly recorded data, making a nice problem.
I can achieve what I want, but only by converting ave from series to dictionary, then to dataframe. Perhaps in this case "good enough" should be let alone.
End Edit
dummy = dummy.append(ave)

Answer 1

It sounds like what you are looking for is pd.rolling_mean:
import pandas as pd
import numpy as np
np.random.seed(1)
# Build example dataframe
df = pd.DataFrame(data=np.random.rand(10,3), index=None, columns=list('ABC'))
print(df)
# A B C
# 0 0.417022 0.720324 0.000114
# 1 0.302333 0.146756 0.092339
# 2 0.186260 0.345561 0.396767
# 3 0.538817 0.419195 0.685220
# 4 0.204452 0.878117 0.027388
# 5 0.670468 0.417305 0.558690
# 6 0.140387 0.198101 0.800745
# 7 0.968262 0.313424 0.692323
# 8 0.876389 0.894607 0.085044
# 9 0.039055 0.169830 0.878143
dummy = pd.rolling_mean(df, window=3).dropna()
print(dummy)
yields
A B C
2 0.301872 0.404214 0.163073
3 0.342470 0.303837 0.391442
4 0.309843 0.547624 0.369792
5 0.471245 0.571539 0.423766
6 0.338436 0.497841 0.462274
7 0.593039 0.309610 0.683919
8 0.661679 0.468711 0.526037
9 0.627902 0.459287 0.551836

Answer 2

Here's another way with a datelike index.
In [67]: df = pd.DataFrame(data=np.random.rand(10,3), index=None, columns=list('ABC'))
In [68]: df
Out[68]:
A B C
0 0.417022 0.720324 0.000114
1 0.302333 0.146756 0.092339
2 0.186260 0.345561 0.396767
3 0.538817 0.419195 0.685220
4 0.204452 0.878117 0.027388
5 0.670468 0.417305 0.558690
6 0.140387 0.198101 0.800745
7 0.968262 0.313424 0.692323
8 0.876389 0.894607 0.085044
9 0.039055 0.169830 0.878143
This is a regular index, but irregular in time (or at least pretend)
In [69]: df.index=date_range('20130101 09:00:58',periods=10,freq='s')
In [70]: df
Out[70]:
A B C
2013-01-01 09:00:58 0.417022 0.720324 0.000114
2013-01-01 09:00:59 0.302333 0.146756 0.092339
2013-01-01 09:01:00 0.186260 0.345561 0.396767
2013-01-01 09:01:01 0.538817 0.419195 0.685220
2013-01-01 09:01:02 0.204452 0.878117 0.027388
2013-01-01 09:01:03 0.670468 0.417305 0.558690
2013-01-01 09:01:04 0.140387 0.198101 0.800745
2013-01-01 09:01:05 0.968262 0.313424 0.692323
2013-01-01 09:01:06 0.876389 0.894607 0.085044
2013-01-01 09:01:07 0.039055 0.169830 0.878143
Take every 3s of data (whether its their or not) and mean it (or you could do fancier if you want). Their are a bunch more options (e.g. which side to include, where to put the labels etc, see here
In [71]: df.resample('3s',how=lambda x: x.mean())
Out[71]:
A B C
2013-01-01 09:00:57 0.359677 0.433540 0.046226
2013-01-01 09:01:00 0.309843 0.547624 0.369792
2013-01-01 09:01:03 0.593039 0.309610 0.683919
2013-01-01 09:01:06 0.457722 0.532219 0.481593