lexical function tree kernels do not implement lambda correctly
Closed this issue · 1 comments
bethard commented
Original issue 414 created by ClearTK on 2015-02-03T19:53:50.000Z:
The lambda parameter for a tree kernel typically limits the weight of very large structures by discounting the number of common subtrees to a value between 0 and 1. The new feature of allowing partial matches of lexical items needs to be multiplied by lambda to be correct in practice. This was missed in testing by only using lambda=1.0. New tests will check for case where lambda != 1.0.