It constructs DMatrix inside training, hence the time interval is actually a sum of data construction and training. It's not required to fix it since it's difficult to separate the two procedures and we might not run it again in the future. This is just a note for anyone who might want to run it.
For comparing XgbGPUHistDaskOldAlgorithm and XgbGPUHistDaskAlgorithm, one can put the DaskDMatrix construction inside timer to make sure it's a fair play, while giving up the accuracy of training time.
It constructs
DMatrixinside training, hence the time interval is actually a sum of data construction and training. It's not required to fix it since it's difficult to separate the two procedures and we might not run it again in the future. This is just a note for anyone who might want to run it.For comparing
XgbGPUHistDaskOldAlgorithmandXgbGPUHistDaskAlgorithm, one can put theDaskDMatrixconstruction inside timer to make sure it's a fair play, while giving up the accuracy of training time.