Fig. 1.
The runtime performance of SIMBSIG compared to the scikit-learn baseline averaged over 10 runs. We tested K-Means (a), a KNN search (b) and PCA (c). Both, in-core and ooc scenarios were computed, whereas the maximum feasible dataset size for in-core was 105 datapoints. For ooc computations, there were no memory restrictions and, therefore, a wider range of dataset sizes could be explored. The batch size was 104. All runs were performed on an Intel i7 10700K CPU and 32GB RAM, the GPU was an Nvidia GeForce RTX3080. Stopping criteria for K-Means were set to default values