search

WladimirSidorenko / CRFSuite

Tree-Structured, First- and Higher-Order Linear Chain, and Semi-Markov CRFs
Other
44 stars 11 forks source link

Is tree structure known during prediction? #7

Closed usptact closed 8 years ago

usptact commented 8 years ago

During training time, one must provide training examples - trees. What about at the prediction stage? Should the tree be predefined? Will an algorithm only label the nodes of the tree?

I am having a problem to map a sequence to a tree. My training set is comprised of tuples - (sequence,tree). In the prediction stage, I am only given sequences. The algorithm should infer the structure of the tree and also label the nodes. Finally, the tree in the output can have more nodes than elements in the input sequence. Can you suggest something here?

Thanks

WladimirSidorenko commented 8 years ago

For the tree-CRFs, the tree structure of an input instance should be known in advance both during training and testing. From your description, I assume that you need some kind of a parser rather than sequence labeler.

usptact commented 8 years ago

Thanks @WladimirSidorenko ! This is what I thought too! Closing this issue.