Syntaxnet: Where is the baseline_eval.py

[felixgwu]

I am wondering where the baseline_eval.py mentioned in the CoNLL 2017 document is.
I searched through the repo, but couldn't find it.

[concretevitamin]

/cc @calberti @bogatyy

[djweiss]

Looks like that's a dangling reference -- we will fix that and add more up-to-date evaluation instructions. In the meantime, please see evaluator.py, segmenter-evaluator.py, and parser-to-conll.py in the dragnn/tools directory to run evaluations.

[martinpopel]

The official CoNLL2017 evaluation script can be dowloaded from http://universaldependencies.org/conll17/evaluation.html. Note that unlike the above-mentioned scripts, it handles different tokenization and multi-word tokens.

[hans]

Hi @djweiss , I'm not able to use the evaluator.py script with the distributed CoNLL 2017 baseline model. (I've tried loading the English and Spanish models; no difference.)

I've reposted the error on a separate GH issue: #1355 (comment)

It seems like the models might be mismatched with the public eval code. Or maybe I'm doing something wrong in the invocation?

I appreciate your help, thanks!