An Adaptable Framework for Deep Adversarial Label Learning from Weak Supervision