Is this as simple as using the new TMLE.gradient_and_estimate defined with weights instead of the classical one ? Could serve as a good comparison withe the more difficult to implement CCW-TMLE.