How to get the prediction label for the shape of [pred_lens, 1].

[QHofmann]

he predicted result in the project has a pred shape of win_size * (predulen- win_size）/ step +1)， It seems that each time step uses 100 (win_size) steps for prediction, but the sliding window cannot guarantee that there will be so many time steps in each test set for prediction