Figure 12.

The procedure for detecting and characterizing events. (A) Raster-gram of the data in Figure 11A for an amplitude of 70% and the time segment between 650 ms and 850 ms. (B) Distance matrix for the trials shown in panel A. The pixel on the ith row and jth column represents the Victor-Purpura distance between the spike train on trial i and the one on trial j using the selected value for the temporal resolution parameter q. The range from small to large distances is mapped onto a gray scale going from dark to white. (C) The heuristic for determining q is based on the distribution of the elements of the distance matrix, referred to as the distances. We show (black curve, left-hand-side scale) the dCV_d and (gray curve, right-hand-side scale) the entropy of the distances as a function of q. The coefficient of variation was calculated as the ratio of the standard deviation of the distances over their mean at q-values whose log₁₀ values were uniformly distributed. The dCV_d is the difference between consecutive CV_d values and thus corresponds to a logarithmic derivative. The entropy is obtained by binning the distances and using the standard p log p expression (see section 3.4). The q value chosen by the heuristic is the mean of the q value at which the entropy has a maximum and the location of the deepest trough in the dCV_d that occurs after the highest peak (dashed vertical line). (D) The heuristic for determining the number of spike patterns. The FCM algorithm is used to find N_c clusters in the data, after which the gap statistic G(N_c) is computed. Each cluster is hypothesized to be a spike pattern. The gap statistic measures the reduction of within-cluster variance achieved by clustering relative to a similarly clustered surrogate data set with points uniformly distributed in the hypercube spanned by the range of the original data. The discrete derivative of G is dG(N_c) = G(N_c) − G(N_c − 1). We show the (black curve, left-hand-side scale) G(N_c) and (gray curve, right-hand-side scale) dG(N_c). The error bars are the standard deviation obtained across 50 surrogates. The value of N_c chosen by the heuristic is the one for which dG is maximal (asterisk). (E) Rastergram and (F) distance matrices with the trials reordered according to their cluster membership. The horizontal lines in panel E separate the clusters, and the vertical gray bands are the events. On two instances, events common to more than one cluster were merged, as indicated by the arrows.