this-and-that has stuff worth sharing.
1-basic-use-xgboost gives a quick introduction on how to use xgboost with CPU or GPU. It uses the diamond dataset in seaborn. It is assumed that you have your environment setup for the XGBoost. I suggest checking these sites for the setup: the official XGBoost website, this article. Especially, the latter one I followed for the GPU setup.
2-ordinal-or-not continues from 1, but focusing on some hypothesis testing regarding the 'cut' variable. It demonstrates how you can represent a variable as nominal or ordinal.
3-reconstruction-with-xgboost introduces a reconstruction module for feature engineering before a predictive model training with xgboost. The latent features from the reconstructor's encoder, which has smaller dimension than the original/physical features, can be used instead. This notebook demonstrates the workflow.