About the evaluate method

[linWujl]

It seems there is a bug in the evaluate function? The predict results have been convert to their source tagid, but in the test dataset, target tags are index, so the precision is always 0 in my result.