Currently FTA only counts uniqueness where the cardinality of the set is less than maxCardinality (12,000 by default). So assuming the number of distinct elements in the set is less than 12,000 which is typically highly likely then the uniqueness percentage (and hence the uniqueness count) should be perfect. Once the number of distinct elements exceeds the defined maximum then FTA simply reports -1 to indicate it does not know the answer.
The obvious solution to this problem is to use something like Hyperloglog (see https://en.wikipedia.org/wiki/HyperLogLog) - note: need to use a variant that supports merging. This would then allow us to implement an approx_count_distinct and hence generate this uniqueness metric even if the cardinality of the set is high. This would be extremely close but not perfect.
Currently FTA only counts uniqueness where the cardinality of the set is less than maxCardinality (12,000 by default). So assuming the number of distinct elements in the set is less than 12,000 which is typically highly likely then the uniqueness percentage (and hence the uniqueness count) should be perfect. Once the number of distinct elements exceeds the defined maximum then FTA simply reports -1 to indicate it does not know the answer.
The obvious solution to this problem is to use something like Hyperloglog (see https://en.wikipedia.org/wiki/HyperLogLog) - note: need to use a variant that supports merging. This would then allow us to implement an approx_count_distinct and hence generate this uniqueness metric even if the cardinality of the set is high. This would be extremely close but not perfect.