Zach Deane-Mayer

I would also like to see this improved! The current version of the package makes my `WITH` statements unreadable

You can add a dummy package.json inside the github action: ``` - name: Dummy package.json so actionlint works run: | rm package.json echo '{ "name": "dummy-package", "version": "1.0.0", "private": true...

I changed the defaults for the trainControl in 4.0. Instead of using 25 fold bootstrap sampling (the caret default), I'm using 5 or 10 fold CV. It's a little convoluted,...

Here's an example: ```R # Fit the old model devtools::install_github("zachmayer/[email protected]") set.seed(42) models_2.0.3

trainControl sets random seeds for each csv fold. make sure your new code and old code have the same number of folds. get caretList working first, then work on caretStack!

Good luck!

What happens with repeats=1? On Mon, Oct 14, 2024 at 6:28 AM BioinfoMonzino ***@***.***> wrote: > Dear Zack, I don't want to bore you again or be inappropriate, but I...

Can you simplify your code more? If you remove all but one model in the list, do the results still differ? If you remove the stack and just look at...

Ok, the first issue (repeats=10, 45 vs 450 observations) is because of this feature I added: `Allow ensemble of models with different resampling strategies, so long as they were trained...

In your second example, I suspect the issue is still in the trainControl. Could you also try: 1. Setting the `metric` in caretList/caretStack to `ROC` 2. Use `twoClassSummary` in the...