Closed normalOne95 closed 5 years ago
Yes, that kernel only outputs spans with high unary scores, but you shouldn't need to touch that at all.
Without changing any modeling code, the spans with high unary scores (specified by this cutoff: https://github.com/kentonl/e2e-coref/blob/master/experiments.conf#L29) should be available to you via top_span_starts
and top_span_ends
at https://github.com/kentonl/e2e-coref/blob/master/predict.py#L32.
Hi @kentonl ,
Apart from the coreference clusters, I also want to extract mention spans that are not present in any cluster (i.e. singletons) but had high unary scores. How can I do this? Does the kernel you wrote (coref_kernels.so) only extract the spans with high unary scores?
Thanks,