Splitting the dataset into train and test parts