The true value of predictions, however, is not about validating the robustness of our model; it's about making predictions on our prediction dataset, in our context, getting survival predictions on this new list of passengers contained in the held-out dataset. The rows in the results file follow the exact same order as the rows in the prediction file. We can put side by side the first rows of the held-out file and the first rows of the results file, and see that the survived and the trueLabel columns are identical:
