Performance scaling with sample size and network size.

(A) Computational processing time in log scale of our method (purple) and Carrillo et al [16] (blue), as a function of sample size T (bins). Synthetic data generated using N = 300, medium density (see Methods for details) and 35 neurons per ensemble. Dots are averages, and shaded regions are ± 1 standard deviation of 100 repetitions. (B) Number of detected ensembles. The dashed red line is the real Number of ensembles. (C) Correlation between the detected and true global ensemble sequence. (D) Average correlation between detected and true individual ensemble sequence, and (E) between detected and true ensemble core-cells. (F) Number of detected ensembles as a function of network size. The Number of core-cells per ensemble corresponds to the 35% of the network size. (G) Same as C, but as a function of network size. Due to computational cost, only our method was evaluated as a function of network size T = 5000.