Agree on an approach for handling units in `.csv` files

[irm-codebase]

One of the last steps in establishing interfacing between a module and the 'outside' is units metadata for .csv files.

This is the only file type we handle that has this issue, since netCDF files have units as standardized metadata.

Based on several tests, I believe the best approach is to add a second 'header' that states the unit of each column, using No Unit to specify columns with text values, ratios, etc.

Reasoning:

• Removing a second header row is easy in pandas
• Fits well with the 'tidy' dataframe approach.
• It produces way less data overhead than adding an extra column in the case of timeseries.
• calliope v0.7 is able to skip rows easily, so it avoids extra processing on the modelling side.
• Allows module devs to easily use pint for unit conversion, if they choose to.

Examples

Picture this table:

attribute,country,vehicle_type,vehicle_subtype,carrier,year,TotalEnergyConsumption
units,No Unit,No Unit,No Unit,No Unit, years, ktoe
index,,,,,,
0,DEU,Powered two-wheelers,Gasoline engine,Gasoline,2000,476.0664153213859
1,DEU,Powered two-wheelers,Gasoline engine,BioGasoline,2000,0.0
2,DEU,Passenger cars,Gasoline engine,Gasoline,2000,29431.27094818872
3,DEU,Passenger cars,Gasoline engine,BioGasoline,2000,0.0
4,DEU,Passenger cars,Diesel oil engine,Diesel,2000,8255.653519490721

attribute	country	vehicle_type	vehicle_subtype	carrier	year	TotalEnergyConsumption
units	No Unit	No Unit	No Unit	No Unit	years	ktoe
index
0	DEU	Powered two-wheelers	Gasoline engine	Gasoline	2000	476.0664153213859
1	DEU	Powered two-wheelers	Gasoline engine	BioGasoline	2000	0.0
2	DEU	Passenger cars	Gasoline engine	Gasoline	2000	29431.27094818872
3	DEU	Passenger cars	Gasoline engine	BioGasoline	2000	0.0
4	DEU	Passenger cars	Diesel oil engine	Diesel	2000	8255.653519490721

Loading and removing the second header:

If you do not want to use any fancy libraries to handle units, cleaning the data is trivial:

data = pd.read_csv("tmp/test2.csv", header=[0, 1], index_col=0)
data.columns = data.columns.droplevel("units")
data.head()

Automatic unit conversion with pint

If you want to be fancy (and lazy), you can just as easily use pint to do all the unit heavy lifting for you.

data = pd.read_csv("tmp/test2.csv", header=[0,1], index_col=0)
data = data.pint.quantify(level=-1).head()
data.head()

data.dtypes

data['TotalEnergyConsumption'].pint.to_base_units()

[jnnr]

I like this approach. Just a tiny request - why not choose a space-free signifier for unitless quantities? E.g.: unitless or no_unit?

Apart from this, I agree with what you propose.

[irm-codebase]

@jnnr I like a the idea, however...

The problem is that pint libraries use No Unit (in pandas) and no unit (in xarray) as standard for unitless cases. I have not tested if capitalisation matters (hopefully not).

You can re-specify it, but using our own standard would add boilerplate, and introduce bugs if said boilerplate is not there.
So even if I dislike it, sticking to the pint naming would save us trouble, I think.

[irm-codebase]

Pinning @brynpickering, as he might find this discussion interesting.