The current implementation of the sumtree (sumtree.py) gives a divide by zero when computing the min value, which is needed for the max wi division in the importance sampling calculation.
The np.min is computed over all nodevalues, and if the tree buffer data is just initialized with less than capacity elements, then at least one entry is zero, thus giving a divide by zero warning.
The current implementation of the sumtree (sumtree.py) gives a divide by zero when computing the min value, which is needed for the max wi division in the importance sampling calculation.
The np.min is computed over all nodevalues, and if the tree buffer data is just initialized with less than capacity elements, then at least one entry is zero, thus giving a divide by zero warning.