Weird behavior on group by TimeGrouper followed by agg

[victorpoluceno]

I'm having problems with groupby with TimeGrouper followed by agg when DataFrame has two lines with unique values.

from datetime import datetime

import pandas as pd
import numpy as np

result = np.empty(3, dtype=[('name', 'object'),
                            ('value', 'int64'),
                            ('date', 'datetime64[us]')])
# doesn't work
result[0] = ('a', 1, datetime.now())
result[1] = ('b', 1, datetime.now())

# this works
#result[0] = ('a', 1, datetime.now())
#result[1] = ('a', 1, datetime.now())

# this also works
#result[0] = ('a', 1, datetime.now())
#result[1] = ('b', 1, datetime.now())
#result[2] = ('c', 1, datetime.now())

df = pd.DataFrame.from_records(result, index='date')
df = df[df['name'].map(lambda x: not x is None)] # filter blank series

group_list = ['name', pd.TimeGrouper(freq='%sS' % 60)]
grouped = df.groupby(group_list)
result = grouped.agg(['sum', 'count'])
print result.to_records()

The main idea is to group by name and date by resample it into 60 seconds intervals. On this example the desired output would be something like this:

[('a', Timestamp('2014-11-11 20:22:00'), 1, 1)
 ('b', Timestamp('2014-11-11 20:22:00'), 1, 1)]

But it returns this:

[(<pandas.tseries.resample.TimeGrouper object at 0x2fbea10>, 'b', 1, 1, 1)
 ('name', 'a', 1, 1, 1)]

Sometimes agg ignores data if there is nothing to aggregate on, this doesn't seems right to me.

show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 2.6.6.final.0
python-bits: 64
OS: Linux
OS-release: 2.6.32-431.29.2.el6.x86_64
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

pandas: 0.15.1
nose: None
Cython: 0.21
numpy: 1.9.1
scipy: None
statsmodels: None
IPython: None
sphinx: None
patsy: None
dateutil: 2.2
pytz: 2014.9
bottleneck: None
tables: None
numexpr: 2.4
matplotlib: None
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
rpy2: None
sqlalchemy: None
pymysql: None
psycopg2: None

[jreback]

this is a dupe of #8542

the issue is that the freq is not valid (and hence is None) for only 2 samples (you need at least 3 to have a frequency). I think its possible to relax this, esp when resampling.

[jreback]

pull-requests are welcome to fix!

[victorpoluceno]

@jreback thanks. I'm going to give it a try and see if I can fix it.

[jreback]

awesome!
lmk if u need help