Figure 7:

Runtimes for different distance metrics. Runtimes for three different distance metrics (Cosine Similarity, Bray-Curtis, and Jensen-Shannon) in Libra with 40 samples of input (372 GB in total). Libra was performed with default parameters (k = 20 and Logarithmic weighting). Runtimes were averaged over three runs. An inverted index was reused for all three distance metrics because the inverted index Libra constructs are independent of the distance metrics. Cosine Similarity took the shortest runtime among the three metrics while Jensen-Shannon took the longest. Jensen-Shannon took almost twice as long as Cosine Similarity because it requires more mathematical computations. Because of its fastest runtime, Cosine Similarity is used by default in Libra.