In this repo, there are two Google Colab notebooks that contain identical tasks. The only difference is one uses pandas / scikit-learn on CPU to preprocess, engineer features and train models while the other uses the equivalent objects and methods in cudf / cuml on GPU to derive the same solutions. The main aim was to compare performance of these two approaches.
amanlai/sales-prediction
Sales prediction using pandas/scikit-learn on CPU vs cudf/cuml on GPU
Jupyter Notebook