Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Question: Lenta Data #198

Open
AllardJM opened this issue Oct 6, 2022 · 1 comment
Open

Question: Lenta Data #198

AllardJM opened this issue Oct 6, 2022 · 1 comment

Comments

@AllardJM
Copy link

AllardJM commented Oct 6, 2022

Ive found that uplift models can be quite fragile but I was surprised experimenting with the Lenta dataset how a method like class transformation performs so well (in fact the first decile produces more uplift in terms of responders than the full campaign, even at the volume in the first decile as is) and many other technique (I experimented with S and T learners and several uplift RF) fail to get close or just plain fail. Curious if this is your experience or there is something with this data that you suspect could be happening....

@ElisovaIra
Copy link
Collaborator

@AllardJM hi there! Thank you for sharing your experience. Mostly we faced the same result with the Lenta dataset and the class transformation model. We even won a hackathon with this dataset + this model couple years ago. More to say is that initially an author of the class transformation algorithm report that target split should be half and half: 50% of “target =1”. But in reality target split can be any.

Benchmarks for uplift models are really a place to study.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants