Testing: Recall from the previous step that individual blast data-sets corresponding to a single phone during each blast experiment form a Testing-set and data-sets from rest of the phones form a Training-set. The number of blasts in a blast-experiment typically determine the size of the Testing-Set. For each data-set in a Testing-set, we do Event-separation and Feature-extraction steps similar to the training (as seen in Figure 4), further apply the non-linear transformation. From this we have a 90 dimensional feature-vector X* and our aim is to estimate a 2 dimensional world-state vector wpred .