search

facebookresearch / SimulEval

SimulEval: A General Evaluation Toolkit for Simultaneous Translation
Creative Commons Attribution Share Alike 4.0 International
99 stars 35 forks source link

some problems with evaluation results #44

Open sarapapi opened 1 year ago

sarapapi commented 1 year ago

Dear @xutaima, I noticed that there are two problems in the files generated after the remote evaluation:

  1. the config.yaml always reports both source and target as "speech" even if we explicitly pass the text as the target type (I haven't tried to pass text as source)
  2. if the --computation-aware tag is passed, both the "_CA" metrics and the ideal metrics are reported as computationally aware, while, if the tag is not passed, the ideal metrics are correctly shown (but not the computational-aware ones).

Thanks

xutaima commented 1 year ago

Hi @sarapapi, thanks for reporting! I will make PR for the fix

sarapapi commented 9 months ago

Hi, I noticed also that the "metric" field is empty in the instances.log file. (Still, CA and NCA metrics are the same)