XGBoost binary logistic -> calculate probabilities manually using the coefficients problem

Hi,

I have an issue and googling it did not help so far …

I am using XGBoost Linear Model Learner with objective set to binary logistic for a binary classification task.

My understanding is that the model is a logistic regression. (what is difference between reg:logistic and binary:logistic? · Issue #521 · dmlc/xgboost · GitHub)

The model consists of coefficients + intercept.

But when I “manually” (in excel) try to calculate propabilities, I do not get the same propabilities compared to what the XGBoost Predictor calculates.

Formula I am using is:

exp(sum coefficients incl. intercept) / (1+exp(sum coefficients incl. intercept))

I can perfectly recalculate propabilities manually for a logistic regression model build with Weka Logistic (3.7) or H2O generalized linear model builder.

So unsure what might be the issue.

Thx for any help!

Best,

Mat