Learning pandas by rewriting some number crunching code

stats.py: pandas re-implemetation
stats2.py: pandas implementation, but with native pandas groupby (instead of itertools groupby)
stats_old.py: the original native python implementation
old_driver.py: csv output wrapper for stats_old.py

The stats_old.py code loops over Python lists, creating tuples with relevant
statistical data and appending them to lists for sorting and output.

The stats.py uses pandas and its code is much easier to understand.
Unfortunately the old code runs in 1m17s while the pandas implementation takes
5m32s. I believe this is because of the overhead of creating lots of little
DataFrame objects for each portal (most are under 10kb in size). There is only
one data operation that requires analysis of the entire DataFrame's time series
plus the data sets are pretty small so this project may not be in the sweet
spot for pandas but it was a good learning project.

The stats2.py loads everything out of the database into a 500mb dataframe
(taking 5m40s), uses the pandas groupby instead of the itertools groupby, uses
the same inner loop as pandas.py to calculate stats (2 min) resulting in a
7m40s runtime. It also leaves all columns as the int64 dtypes that are returned
from the sqlite database (stats.py copies everything into appropriately-sized
dtypes).