The following is a portion of the original pandas.DataFrame:

This is part of the DataFrame after going through "pd_to_tensor" and "tensor_to_pd":

Obviously, some NaN are being filled in unintentionally.