🏎 ⚡️ 💯 [Rough] Benchmark across Segmentation Tools, Libraries and Algorithms #68
Conversation
…nto npn-benchmark * 'npn-multiple-lang' of github.com:nipunsadvilkar/pySBD: (22 commits) 🔧 Update setup.py 🎨 cleanup ✨ 💫 Support Kazakh language ♻️ Refactor Processor to get language specific Processor ✨ 💫 Support Deutsch language ✨ 💫 Support Japanese language ✨ 💫 Support Italian language ✨ 💫 Support Greek language ✨ 💫 Support Burmese language ✨ 💫 Support French language ✨ 💫 Support Danish language ✨ 💫 Support Dutch language ✨ 💫 Support Persian language ✅ 🔧 Fix PySBDFactory & limit char_span to English lang only ✨ 💫 Support Polish language ✨ 💫 Support Russian language ✨ 💫 Support Urdu language 🚑 Add Bulgarian language files ✨ 💫 Support Bulgarian language ✨ 💫 Support Armenian language ...
nipunsadvilkar
changed the title
There was a problem hiding this comment.
These benchmarks are good for comparing libraries, but to create a "words per second" benchmark it might be good to just run these on a larger quantity of text, e.g over 10k sentences.
One minor comment on the correctness of the stanza benchmark, but otherwise LGTM
|
@DeNeutoy Can you suggest any dataset to benchmark against? |
|
@nipunsadvilkar Perhaps a book from Project Gutenberg? They have full plaintext books, e.g: This would allow us to also analyse failure cases of the various methods also. |
|
Here is another alg to benchmark - https://github.com/microsoft/BlingFire#python-api-description Blingfire is very fast, but I don't know how good their sbd module is. |
|
@DeNeutoy : Benchmarked |
nipunsadvilkar
changed the title
Segmentation Tools, Libraries and Algorithms: