search

hfshr / distill_blog

Blog and things
https://hfshr.xyz
9 stars 2 forks source link

Data, Code & Coffee: When one model is not enough: Stacking models with {stacks} #6

Open utterances-bot opened 3 years ago

utterances-bot commented 3 years ago

Data, Code & Coffee: When one model is not enough: Stacking models with {stacks}

https://hfshr.xyz/posts/2020-11-30-model-stacking/

kamaulindhardt commented 3 years ago

Thank you Harry,

This example is amazingly clear, interesting and I learned a lot! I am right now in the middle of my own adventures with the TidyModels framework where I have constructed several regression sub-models that I wish to stack into an ensemble model. I have ran into an issue with the stacks() function because several of my sub-models have NEG stacking coefficients, even though they perform very well when I check what sub-models perform best in the finalize() step.. This issue is becoming impossible for me to figure out and I would highly appreciate if you could take a look at the code. Can I sent it to you, together with the data, via mail maybe?

hfshr commented 3 years ago

Hi @kamaulindhardt, thanks for your message! Sure, feel free to share your code/data and I can take a look, no promises I'll be able to help though 😅

costebk08 commented 2 years ago

This is fantastic content thanks for sharing! This does present one minor issue though, which I have seen in every example of stacks, which is that the model stack is created using just the training data. Is there a way to fit the stack to the entire dataset while avoiding redoing everything with the entire dataset? Thanks!