[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:02823] Precisely control the number of leaf nodes in cluster


Hi,
I am trying to comtrol the number of leaf nodes of decision trees by acount other then MSD factor. There are two aspects I mainly concern: nLeaves per tree and total number of leaves in one stream.
I wanted to cluster the context decision tree by a buttom-to-top method. Thus, I tried to reduce the leaves per tree in a similar way as ShrinkTree in DR(HHEd cmd) and reduce total leaves per stream in a similar way as the training of xforms in adaptation, i.e. similar leaves share a prob.
In practise, I tried to print the shrinked tree at the beginning of Dec2Reg to a file named "dec.tree.shrinked". However, the indices of the nodes are completely a mess. Why?
Is there any suggestions how I can do this? Thanks all!
--
Na Xingyu
Beijing Institute of Technology