next up previous
Next: Discrete variables of arbitrary Up: The ACCL algorithm Previous: Running time

Memory requirements

Beyond the storage for data and results, we need ${\cal O}(N_C)$ space to store the co-occurrences and the lists $C(u),\overline{V}_0(u)$ and ${\cal O}(n)$ for $L, vheap$ and the Kruskal algorithm. Hence, the additional space used by the ACCL algorithm is $
{\cal O}( s^2N + n )$.

Journal of Machine Learning Research 2000-10-19