TODO

--- TODO ----
Short term before publishing:

- generate instructions for installation.
- generate instruction for testing that installation is OK.


In unittest: 
- check that minimum requirements are installed 
  (R with MLDS package, rpy2)
  
- parallel execution:
  - check installation of snow package 
  - check access to localhost using 'username'
  
  - do a simple test of snow package load, and run a simple calculation


- check if it runs in non-linux OS




--- Long term ---

- make implementation cleaner, use rpy2 in all commands calling R
  (instead of passing R commands with subprocess)

- improve documentation, add docstrings to all functions

- eventually implement the class so it does not depend on R. 
For that, one would have to use a GLM solver in python (statmodels?)
I don't know if this is adviced. We will have to check that both
implementations are equivalent numerically. And then, after an update 
on the original MLDS package, we will have to modify ours as well.