Consider a data frame with a date column as an index and three columns x, y and z with some observations. I want to write the contents of this data frame to a .csv file. I know I can use df.to_csv for this, however, I would like to add a second header line with the units. In this example, the desired .csv file would look something like this:
date,x,y,z
(yyyy-mm-dd),(s),(m),(kg)
2014-03-12,1,2,3
2014-03-13,4,5,6
...
This doesn't produce the exact output in your example, but it's close. You can use multi-index columns to store the second header (the units) with the column labels:
>>> import pandas as pd
>>> columns = pd.MultiIndex.from_tuples(
... zip(['date', 'x', 'y', 'z'],
... ['(yyyy-mm-dd)', '(s)', '(m)', '(kg)']))
>>> data = [['2014-03-12', 1, 2, 3],
... ['2014-03-13', 4, 5, 6]]
>>> df = pd.DataFrame(data, columns=columns)
>>> df
date x y z
(yyyy-mm-dd) (s) (m) (kg)
0 2014-03-12 1 2 3
1 2014-03-13 4 5 6
Storing the second header this way allows your columns to keep the correct type (e.g., column x should be an integer type):
>>> df.dtypes
date (yyyy-mm-dd) object
x (s) int64
y (m) int64
z (kg) int64
dtype: object
If you had stored the second header as a row in the DataFrame, your column dtypes would become object, which you probably don't want.
Writing the DataFrame in CSV format produces something very similar to your example:
>>> df.to_csv('out.csv', index=False)
>>> !cat out.csv
date,x,y,z
(yyyy-mm-dd),(s),(m),(kg)
,,,
2014-03-12,1,2,3
2014-03-13,4,5,6
The only difference is the extra line of commas, which is how pandas separates multi-row headers from the actual rows of data. This allows the CSV file to be read back into an equivalent DataFrame:
>>> df2 = pd.read_csv('out.csv', header=[0, 1])
>>> df2
date x y z
(yyyy-mm-dd) (s) (m) (kg)
0 2014-03-12 1 2 3
1 2014-03-13 4 5 6
Note: I found a lot of this information scattered throughout this SO question.
df.loc(), but apparently I get the syntax wrong (I always get the errorValueError: unsafe appending to index of type DatetimeIndex with a key yyyy-mm-dd).