On Learning for Ambiguous Chance Constrained Problems