|
dogi = Doge.objects.filter(train_end_timestamp=last_timestamp) # .order_by('-metric_value') TODO: check if needed |
The TODO here suggests ordering doges in the query. That's probably not necessary because it doesn't matter which order they vote. However, if we don't do weighted voting, then you could instead trim the query to not let the poor performers vote. Just be careful that if it's a hardcoded value, that it doesn't filter out more than half the size of the committee. It's ok to have some weak ones voting because were going for a "wisdom of crowds"
clover/apps/doge/models/doge_train_test.py
Line 181 in 49f1426
The TODO here suggests ordering doges in the query. That's probably not necessary because it doesn't matter which order they vote. However, if we don't do weighted voting, then you could instead trim the query to not let the poor performers vote. Just be careful that if it's a hardcoded value, that it doesn't filter out more than half the size of the committee. It's ok to have some weak ones voting because were going for a "wisdom of crowds"