NaN for features 1009 and 1012 (LM probability of the s/t sentence)

ghpaetzold / questplusplus

Pipelined quality estimation.

49 stars 14 forks source link

NaN for features 1009 and 1012 (LM probability of the s/t sentence) #42

Closed craigastewart closed 6 years ago

craigastewart commented 6 years ago

The above features included in the 'default' blackbox_17 sentence level feature set result in NaN in the output feature vectors, even when using the english and spanish language models provided in the download.

I'd like to include sentence probability as a feature; is there anything I can do to stop this from happening?

Thanks