Figure 6:

Scalability testing for Libra. Runtimes of Libra on four datasets consisting of 10, 20, 30, and 40 samples (total sizes of 93 GB, 186 GB, 279 GB, and 372 GB, respectively). Libra was performed with default parameters (k = 20, Logarithmic weighting, and Cosine Similarity). Runtimes were averaged out over three runs. The total runtime of Libra increased linearly with increased input volume. Both index construction and distance matrix computation showed linearly increased runtimes for the increased input volume. This shows that Libra performs efficiently and scales to input although the number of distances between sample pairs to be computed increases quadratically.