Accelerations

Targeted estimators are expensive by nature, in particular when resorting to cross-validation and collaborative TMLE. However many of the operations can be conducted in parallel. To facilitate this, an acceleration parameter can be provided to the estimation call.

At the moment, only single CPU and multi-threading modes are supported. For a multi-threaded call:

tmle = Tmle(resampling=CausalStratifiedCV())
tmle(Ψ, dataset; acceleration=CPUThreads())

In this case, fitting across multiple folds will happen on all available threads. Similarly the outcome mean and propensity score model will be estimated in parallel.