metadata

license: apache-2.0

neuspell-scrnn-probwordnoise

towards a reliable workaround for the neuspell lib being broken

here, the SclstmChecker() would be used

original usage

this one would use bert checker

!pip install -U neuspell -q

data_folder = "/usr/local/lib/python3.7/dist-packages/neuspell"


import neuspell
from neuspell import *


""" select spell checkers & load """
# checker = SclstmChecker(pretrained=True)
# checker.from_pretrained()
# ^^are the standard method and broken

url = "https://www.dropbox.com/sh/7u160z9gvijnc06/AACf3RDpbFn76vKS-f5rzvxWa?dl=1" #@param {type:"string"}

!wget $url -O mychecker.zip

import os
target_dir = "/content/dl-checker"
os.makedirs(target_dir, exist_ok=True)
!unzip /content/mychecker.zip -d $target_dir

# checker3 = BertChecker()
checker4 = SclstmChecker()
checker4.from_pretrained(target_dir)

checker4.correct("I luk foward to receving your reply")

checker4.correct_strings(
    ["I luk foward to receving your reply", 
     "hooba dooba im like a pooba",
     "were did wendigo goe goating?"]
    )