I am not sure whether you got my point but I am actually trying to say that if you are not able to calculate the number of partitions needed for your dataset (not considering any aggregations at the initial stage) then just pass some higher number and AQE will take care of the rest.