On the example provided here if I run 100 times the estimation procedure for a dataset of size 100 000 samples the variance of my estimates is consistently ~4.5 bigger than the original implementation from the original package. I need to investigate the reason behind this.
Idea suggested by Sjoerd: see if the original package runs more than one iteration of TMLE updates even for this case as it may be reducing variance.
On the example provided here if I run 100 times the estimation procedure for a dataset of size 100 000 samples the variance of my estimates is consistently ~4.5 bigger than the original implementation from the original package. I need to investigate the reason behind this.
Idea suggested by Sjoerd: see if the original package runs more than one iteration of TMLE updates even for this case as it may be reducing variance.