Another nice extension that would be straightforward would be adding CV-TMLE to the package. I've been working on the drtmle package, which implements TMLE estimators with doubly-robust inference. I set up the structure of that package very similarly to survtmle with calls to estimate functions and fluctuate functions. I found that adding wrappers around the estimate functions to cross-validate initial estimates involved only some pretty straightforward bookkeeping. In other words, I think it will be straightforward to implement in survtmle as well. Studying practical performance of CV-TMLE vs. standard TMLE in competing risk setting would be of interest (i.e., could be a short paper).
Implementation should be rather straightforward. It may perhaps be worth using origami when creating options for parallelization (rather than foreach), allowing access to futures.
nhejazi
commented
7 years ago
Another nice extension that would be straightforward would be adding CV-TMLE to the package. I've been working on the
drtmlepackage, which implements TMLE estimators with doubly-robust inference. I set up the structure of that package very similarly tosurvtmlewith calls toestimatefunctions andfluctuatefunctions. I found that adding wrappers around theestimatefunctions to cross-validate initial estimates involved only some pretty straightforward bookkeeping. In other words, I think it will be straightforward to implement insurvtmleas well. Studying practical performance of CV-TMLE vs. standard TMLE in competing risk setting would be of interest (i.e., could be a short paper).