Figure 9. Estimating the optimal parameters for fast search.

Accuracy is defined as the percentage of the number of identical solutions between MRKD tree and an exact search over 1,000 search results. The dashed line marks the parameter selected. (a) For 1,000 search results, the search accuracy versus the number of trees is presented. In total, 128 trees are selected to produce accuracy of ~76%. (b) With 128 trees, the search accuracy for the top 1,000 solutions versus the total number of output solutions for five random synthetic entries is shown. A total of 1,000 output solutions are sufficient to produce an accuracy of ~76%. (c) To apply PCA dimension reduction, the plot of eigenvalues of the data covariance matrix versus dimensions is presented. Those eigenvalues close to zero at dimension above 100 can be removed and the resulting new data set has only 100 dimensions without losing much information.