Olivier Labayle

Implement the C-TMLE: - https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2898626/ - https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6086775/

Tables are eventually transformed to NamedTuples. I don't know to what extent this could be leading to performance issues.

At the moment it is assumed that TMLE will converge in one step (which is the case for our estimands/fluctuations)

From ACIC: https://arxiv.org/abs/2110.14831 Investigate balancing, i.e. estimate 1/p(T|W) instead of propensity P(T|W). Potentially some gains in finite samples. See if I understood the idea correctly, if and how it could...

This is a conditional average treatment effect when the conditionned variable depends on the backdoor adjustment set. - [x] Derive Influence curve - [ ] Derive fluctuation - [ ]...

From ACIC: https://arxiv.org/abs/2112.13398 Potential new feature

- [ ] Check that En[D] is approximately 0, ideas to understand/investigate --> https://github.com/tlverse/tmle3/blob/425e21c9d93d8b318423d8d393925572d6c9eea1/R/tmle3_Update.R#L285 - [ ] Report risk - [ ] Check norm of epsilon should be lower than...

TODO: - [ ] Address: https://github.com/olivierlabayle/TMLE.jl/issues/49 - [ ] Update docstrings - [ ] Update documentation - [ ] Write paper - [ ] Submit For reference: https://joss.readthedocs.io/en/latest/submitting.html#submitting-your-paper and https://github.com/alan-turing-institute/MLJ.jl/tree/dev/paper