raklokesh / Sum-Tree

Sum tree data structure for stochastic priority sampling

Sum tree data structure

Created for the purpose of sampling from discrete priority distributions stochastically. A priority is associated with each sample point and we can sample stochastically according to the sample priorities.

Sum tree class

1. update_leaf - updates the priority value on the leaf and propagates the change in priority value for that leaf up the tree.
2. update_tree - propagates the change in priority value of a leaf up the tree
3. get_priority - returns a priority given a sum value
4. update_allLeaves - runs update_leaf for all the leaves. Can be used to set all leaf priorities at once and propagate them

Sum tree test script

Originally created to debug the Sum tree class. The last section of the script can be used to generate frequency plot for priorities by selecting a large number of sum values in the range. The frequency plot shows that the sum tree returns priority samples in almost linearly increasing frequency with priority value.