Open
Description
Some kind of BN or PGM workload would be good to include in our project. To see the connection, note that the probability distribution of a PGM is equal to the product of the factors, and most operations involve taking a sum or maximum or both over this product (i.e. an einsum). Specifically, this involves:
- Marginal Calculation (i.e. sum over some of the nodes in the graph)
- Maximum A Posteriori Calculation (i.e. max over some or all of the nodes in the graph)
Sparse factors are pretty rare in this setting, so some justification might be necessary when finding datasets. One "easy" route would be to take the workloads from Rina Dechter's UAI competitions (https://uaicompetition.github.io/uai-competition-dev/) then simply sparsify the factors in some manner.