Question about XGBoost bundle speed.

[drei34]

Hi,

I have a question as I am using this model. I varied the batch size to be 1, 20 and 50 and I measure the latency given a batch size in my java code. It seems pMax latency is largest for batch size 1 and I am trying to make sure I understand why. Also, as I vary the batch size, it seems that on average a batch of size 50 takes less time than a batch of size 1. Is this because of threading (inside the bundle)? How do you turn threading off, can you? For example, see below my latency numbers below. Thank you!