Parameterize size of mini-batches

[bowni]

Parameterize size of mini-batches so that distributed learning and weights aggregating phases can be optimized.

[bowni]

@RomainGoussault this issue could go in a next milestone (0.2), I don't think it is needed to present and demo the project.

[bowni]

Done in PR #83