Using GPU to compute statistical features based on PyTorch.
Also compare the results with the features computed by CPU (Numpy).
The return is a pd dataframe with columns: 'feature name', 'feature value gpu', 'feature value cpu', and 'time consumption'.