we use version 0.80, ‘XGBoostClassifier’ to train, ‘auc’ to evaluate, but if set ‘numWorkers’ > 1, the auc result is unstable(test several times with same params, the ‘auc’ result may be different, and not awalys different). if set ‘numWorkers’ = 1 and same params, the ‘auc’ result always same. This problem troubled me many days, could anyone help me?
Xgboost-spark unstable
sounds like random seed for sub-sample is not fixed on Spark Xgboost
we always set subsample=1,a fixed value, and any other suggest?
Was there any solution to the above issue?