フリー問題

問題文

Which pair of settings moves XGBoost training onto the GPU in the current API?

選択肢

1. Set the device parameter to the value naming the accelerator and keep the tree method at the histogram algorithm.
2. Set the objective to a value with an accelerator prefix, so the loss function carries the choice of hardware and the device argument becomes redundant.
3. Set the tree method to the historic accelerator-specific value that combines the device and the algorithm in one string, which remains the documented way of requesting accelerated training because it is the only value the booster recognizes for that purpose.
4. Move the input matrix to device memory and leave the parameters unchanged, so the location of the data decides where the trees are built.