I am running some experiment using XGBoost. I set number of iterations to 1000 and get result - reasonable ones..
Than I tried to do something different:
Instead of run 1000 iterations I would like to run 100 iterations, a this repeat 10 times (total also 1000 iterations). Between this 10steps I setup: target_learn = target_learn - target_fitted (from the step before).
and than target_fitted = target_fitted + prediction
Target_fitted prediction is gain by Cross validation on test sets over all dataset.
I got different results - very poor accuracy.
Than I go even far a run only 5 iterations done 200times... even worse..
The reason why I am trying to do this, that I want to change learning matrix during iterations (trying to do something little bit recurrent).
Any idea why the result is different... it is seems to me, that first iteration in each step is something different, so I have this iteration in model many times, which is killing acurracy...
Other parametrs are same..
or maybe I am missing completele something..? thx
