We also used t-sen to embed the two hybrid data sets described in Methods / Data Sets. The results for HD1 can be seen in Figure 3A and for HD2 in 3B. In this case, the output of the t-sne algorithm provides a clean visualization of the known single unit clusters. Most clusters are fully separated from the other clusters and contain a very small number of spikes that do not belong to their corresponding unit. The spikes shown in black in both sets are not hybrid spikes, but rather those that existed in the original recording, i.e. not ground-truth. For HD1 all clusters have Precision, Recall and F-factor of 0.99. For the HD2 set, clusters 1, 4, 5 and 7 have Precisions, Recalls and F-factors ranging between 0.95 and 0.99. Cluster 6 has 0.91, 0.9 and 0.91 respectively. In the case of cluster 5, DBSCAN did not classify as part of the unit the spikes that expand away from the main group to the right of the cluster. Yet their proportion was small enough to keep the Recall at 0.95. The same applied to cluster 1, which was missing the group of spikes showing at the top of the large not Ground Truth spike grouping (black group at the left of the plot labeled as Not GT). That group of spikes was also small enough in percentage that the cluster still achieved a Recall of 0.96. T-sne failed to fully separate clusters 2 and 3 (see insert of Figure 3B).