How is the running speed?

yzhangcs / parser

:rocket: State-of-the-art parsers for natural language.

https://parser.yzhang.site/

MIT License

829 stars 141 forks source link

How is the running speed? #5

Closed 2020zyc closed 5 years ago

2020zyc commented 5 years ago

hi, how is the decode speed? Any evaluation? Thx.

yzhangcs commented 5 years ago

I didn't evaluate the parsing speed precisely like the paper, so it may not be able to compare with it correctly. The time for training an epoch is about 4m50s, and it takes about 1m12s to evaluate the three datasets (43948 sents). I didn't estimate the time of decoding individually.

2020zyc commented 5 years ago

ok, thx and nice code.

yzhangcs commented 5 years ago

Thanks :)

yzhangcs commented 5 years ago

Recently I implemented the bucket mechanism adopted in tensorflow, which has made the speed twice as fast as I said before, but may result in a slightly loss of performance (95.79 & 94.08, still better than the official code). The implementation is still not perfect, but may give you some inspiration.