Has anyone been able to implement a conditional logistic?

[brebbles]

Thanks for your comments. After thinking about it and playing with it a bit more, I think that the way I described is probably still logically consistent, but I don’t think the intuition for the conditional logit case is as compelling as for the binary case. Since you mentioned cross-validation, I thought I would mention that I am working on a routine which would perform LOOCV on fitted XGBoost models (leaf nodes only, conditional on the structure and splits). Would you happen to know if anything like that exists already?

Hi compleathorseplayer,

Unfortunately I don’t know of any current functions to perform LOOCV on fitted models. I’ve only started looking at creating manual CV functions myself to fill in features that xgBoost seems to fall down on currently - eg using the base_margin doesn’t pass through currently when running a multiclass CV as in here

[compleathorseplayer]

Thanks for your imput.

[brebbles]

Hi compleathorseplayer,

I notice that XGBoost have now implemented Accelerated Failure Time as part of the survival objective - this allows for both left and right censored times so it seems like it is possible to implement a condtional logistic using this XGBoost now.

https://xgboost.readthedocs.io/en/latest/tutorials/aft_survival_analysis.html

There may be an issue with slicing the data matrix for cross-validation though I think they are working on patching that issue (if they haven’t already).

It would be interesting to compare the output of this method with the custom objective/error functions I wrote above and see if one significantly out-performs the other

[AiUser]

Hi Brebble. Have you had any chance to implement CL with xgboost. I had it working but I am unsure of the quality of the results related to the way xgboost bootstrap data within or not the strata.