Analyst: task-balance produces unbalanced training/eval sets

[PetrToman]

Task-balance (used in Workbench, for instance) produces unbalanced training and eval sets:

As depicted, data_balance.csv contains balanced ideal outputs (only two classes in this simplified example, i.e. y=0 or 1), but this balance is not preserved in the next step.

[jeffheaton]

Right now balance is just a single command that is executed once when running a task. It is done once, when commanded, but it is not re-balanced for every other command. To do it the way requested here I would need to make balance to be an option for each command and not a command itself. Which, might be useful. I am moving this to an enhancement for future release.