TMondain
commented
2 years ago The following plots are only going to be for the uptake = 0.1 results, just to keep things simple. It does seem like there's the same problem across different uptake values.
The distribution of the change in evaluation metrics is incredibly narrow.
It seems like the distribution of all 'change from initial models' have outliers, irrespective of the evaluation metric being used. It's also not always the same communities that have the outliers.
This is an expansion of issue #11, which was focused on a single community. It's becoming apparent that outlier communities are pretty common across all of the adaptive sampling rounds, so thought it would be good to have all the investigations in one place.