How to include the confidences in the forecasting results?

[Cram3r95]

At this moment I am including my multimodal predictions as following:

pred_traj_fake_global_aux = pred_traj_fake_global.squeeze(0).view(ARGOVERSE_NUM_MODES, PRED_LEN, 2)
output_all[seq_id] = pred_traj_fake_global_aux.cpu().numpy()

Nevertheless, I would like to know how to include the confidences in this array since ARGOVERSE evaluates the multimodal prediction according to these confidences.