Right now, constraints are ignored in HyperSHAP. This may lead to scores for hyperparameters that are due to performance predictions on invalid hyperparameter values and thus not reachable in practice.
Consider constraints in search-based explanation games.