Currently, we can only build ML databases from sampler output with stat-files. However, the sampler can write objectives to the JSON file like the optimizer.
