search

yubin-park / bonsai-dt

Programmable Decision Tree Framework
https://yubin-park.github.io/bonsai-dt/
34 stars 6 forks source link

AlphaTree risk pyramid #4

Open martinmocko opened 5 years ago

martinmocko commented 5 years ago

Hi Yubin,

I would like to know how I can create a risk pyramid similar to the one provided in your work in Figure 5. The steps necessary to do so, after fitting an AlphaTree, aren't very clear to me. I know it will probably be calculated by using the current ratio of classes but I'm not sure if also the predicted y_label comes into play somehow.

Looking forward to hearing from you!

yubin-park commented 5 years ago

Thanks, @zerodayexploit ACDC is not currently implemented in this project, but will add. When I add, I will also put some example code to create the risk pyramid presented in the paper. Will leave a message here when things are done. Thanks so much.

martinmocko commented 5 years ago

Ah, sorry @yubin-park , I somehow assumed you had the ACDC implemented since in the AlphaTree code I found a comment that says it's closer to the ACDC implementation. Maybe I could be of help concerning the implementation, if you could give me some pointers about what needs to be changed/implemented.

yubin-park commented 5 years ago

@zerodayexploit To implement ACDC, you would need to grow only one branch, increasing the ratio of positive samples at each iteration. The current implementation of bonsai may not be well suited for doing this task, unfortunately. This may need some change in the core library. Let me get back to you in a few weeks.