What is the mechanism of using param 'scale_pos_weight' in xgboost?

Question

My dataset has 90% negative samples and 10% positive samples which is very imbalanced. I try to use the parameter of scale_pos_weight and set it as 9. What is the mechanism of this param do. I am curious about what it actually means: does it mean repeat the positive samples 9 times? Or everytime pull out 1/9 samples of negative samples and train the model in many times. besides, if I have a dataset whose negative samples just a little more than the positive ones, should I need to specify the parameter again?

Answer 1

scale_pos_weight in xgboost is just used to multiply the weights as is source code. There is no subsampling done based on this parameter.