Closed lendle closed 10 years ago
I don't want to merge this to master yet, so I'm closing this pull request. I have created a branch on my repo called "rare_outcomes" if you'd like to see it.
Hi Sam,
Ok, sounds good. I'm still on vacation until Jan 13 but I did some work on the variance estimation under sparsity piece a few weeks ago. It's in a branch if you want to look at it.
Josh
From: lendle notifications@github.com To: joshuaschwab/ltmle ltmle@noreply.github.com Sent: Sunday, December 29, 2013 12:17 PM Subject: Re: [ltmle] Adding boundedlogistic(), Qbounds option to ltmle (for rare outcomes), and tests (#7)
I don't want to merge this to master yet, so I'm closing this pull request. I have created a branch on my repo called "rare_outcomes" if you'd like to see it. — Reply to this email directly or view it on GitHub.
Hi Josh,
I added some variance estimators to the rare_outcomes branch. It looks like you started doing something similar in your branch, so maybe this will be helpful. They are basically counting on that there is only one regime and IC and Qstar are one dimensional. Afaik that should always be the case for ltmle(), but the way it's set up now is not great. Since I'm not familiar with how a lot of the parts in the j in length(nodes$LY):1 loop in FixedTimeTMLE work, I'm not sure how best to integrate everything. Maybe we can discuss it when you get back?
Sam
On Thu, Jan 2, 2014 at 5:25 PM, joshuaschwab notifications@github.comwrote:
Hi Sam,
Ok, sounds good. I'm still on vacation until Jan 13 but I did some work on the variance estimation under sparsity piece a few weeks ago. It's in a branch if you want to look at it.
Josh
From: lendle notifications@github.com To: joshuaschwab/ltmle ltmle@noreply.github.com Sent: Sunday, December 29, 2013 12:17 PM Subject: Re: [ltmle] Adding boundedlogistic(), Qbounds option to ltmle (for rare outcomes), and tests (#7)
I don't want to merge this to master yet, so I'm closing this pull request. I have created a branch on my repo called "rare_outcomes" if you'd like to see it. — Reply to this email directly or view it on GitHub.
— Reply to this email directly or view it on GitHubhttps://github.com/joshuaschwab/ltmle/pull/7#issuecomment-31499600 .
Still need to implement alternate variance estimators, but this allows Qbar to be bounded, and can improve performance for rare outcomes.
The boundedlogistic function uses optim()'s LBFGS method for optimization, but that could be changed later if something works better.