Can we use other ensemble decision tree? about dtreeviz HOT 12 CLOSED

hi @hcho3 , @martin-loetzsch , @H4dr1en
we've just merged xgboost implementation into master. Before making a new release, would be helpful to have a feedback from your sides.

You can track more comments from #83 and #91

For more details about visualizations, you can take a look at https://github.com/parrt/dtreeviz/blob/master/notebooks/dtreeviz_xgboost_visualisations.ipynb

Thanks, Tudor.

from dtreeviz.

I filed a pull request in the XGBoost repository to mention dtreeviz: dmlc/xgboost#6013. Thank you for the great work!

from dtreeviz.

@tlapusan Mainly we want to put up a self-contained demo script that can be automatically run by our CI system. That way, any future changes of XGBoost won't break dtreeviz accidentally.

from dtreeviz.

@hcho3 Have you been able to write that wrapper class for XGBoost? And if so, can you share it?

from dtreeviz.

closing this as we have xgboost and @tlapusan is working on lightgbm

from dtreeviz.

Not currently. You would just have to get ShadowDecisionTree wrappers for those trees. Note that boosting trees are usually very shallow; not sure it'll help but try it out!

from dtreeviz.

Thank you for the response

from dtreeviz.

This is a really cool project. It would be interesting to see XGBoost integrated into dtreeviz.

You would just have to get ShadowDecisionTree wrappers for those trees.

So I’d just need to define a new class that looks like ShadowDecTree? I have some experience parsing XGBoost models; let me see if I can make a working pull request.

from dtreeviz.

@hcho3 yep, would be nice. I would consider a PR. :)

from dtreeviz.

Can we re-open? A support for XGBoost would be very interesting!
I am pretty sure it is feasible using the trees_to_dataframe function to get a dataframe description of the trees:

from dtreeviz.

Ah. ok. maybe somebody could submit a PR.

from dtreeviz.

Thanks @hcho3 for your support. Is there anything we can help with the demo, or is something internal to xgboost team?

from dtreeviz.